Univerzltet erne Gore ElEKT'ROTEHNICKI FAKlJlTE1' …...Univerzltet erne Gore v ElEKT'ROTEHNICKI FAKlJlTE1' ucc 81000 Podqorioa, OZ. Vasingtona bb, tel. t.R. 510-255-51, PIS: 02016702

Univerzltet erne Gore v

ElEKT'ROTEHNICKI FAKlJlTE1' ucc 81000 Podqorioa, OZ. Vasingtona bb, tel.

t.R. 510-255-51, PIS: 02016702 245 fa'c (020) 245 87'3 PDV: 30/31-03951·6 Univerzitet erne Gore

UNIVERZITET CRNE GORE

- Centru za doktorske studije -

- Senatu -

OVDJE

U prilogu dostavljamo Odluku Vijeca Elektrotehnickog fakulteta sa sjednice od 23.01.2018. go dine i obrazac D2, sa pratecom dokumentacijom, za kandidatkinju mr Siditu Duli, na dalji postupak.

-, ucc

UNIVERZITET eRNE GORE Obrazac D2: Ispunjenost uslova doktoranda

ISPUNJENOST USLOVA DOKTORANDA

OPSTI PODACI 0 DOKTORANDU Titula, ime, ime roditelja, MSc Sidita Duli prezime Fakultet Elektrotehnicki fakultet Podgorica Studijski program Doktorske studije elektrotehnike Broj indeksa 1/2009

NAZIV DOKTORSKE DISERTACIJE

Na sluzbenorn jeziku Pristup paralelizaciji estimatora parametara Weibullove distribucije sa primjenama

Na engleskom jeziku An approach for the parallelization of the Weibull distribution parameter estimators with applications

Naucna oblast Racunarstvo

MENTOR/MENTORI Univerzitet Crne

Gore Automatika i Mentor Prof. dr Bozo Krstajic i Elektrotehnicki fakultet Podgorica racunarstvo

Crna Gora KOMISIJA ZA PREGLED I OCJENU DOKTORSKE DISERTACIJE

Univerzitet Crne Gore Digitalni sistemi i Prof. dr Milutin Radonjic Elektrotehnicki fakultet Crna Gora informatika

Univerzitet Crne Gore Automatika i Prof. dr Bozo Krstajic Elektrotehnicki fakultet racunarstvo Crna Gora

Univerzitet u Beogradu Racunarska tehnika i Doc. dr Slavko Gajin Elektrotehnicki fakultet Beograd informatika Srbija Datum znacajni za ocjenu doktorske disertacije Sjednica Senata na kojoj je data saglasnost na ocjenu teme i 27.06.2013. godine kandidata Dostavljanja doktorske disertacije organizacionoj jedinici i Saglasnost mentora: 12.12.2017. saglasanost mentora Predaja disertacije: 18.12.2017. Sjednica Vijeca organizacione jedinice na kojoj je dat prijedlog za imenovanje komisija za pregled i oCJenu 23.01.2018. godine doktorske disertacije

ISPUNJENOST USLOVA DOKTORANDA U skladu sa Clanom 38 pravila doktorskih studija kandidat je dio sopstvenih istrazivanja vezanih za doktorsku disertaciju publikovao u casopisu sa SCI/SCIE liste kao prvi autor. Spisak radova doktoranda iz oblasti doktorskih studija koje je publikovao u cascpisima sa (upisati odgovarajucu listu)

Obrazac D2: Ispunjenost uslova doktoranda 1/3

-, UCG

UNIVERZITET eRNE GORE Obrazac D2: Ispunjenost uslova doktoranda

Rad u casopisu sa SCI liste:

[1] S. Duli, B. Krstajic, "Parallel Implementation of the Weibul1 Distribution Parameters Estimator", The Journal of Environmental Protection and Ecology crEPE), ISSN 1311- 5065, VoLl5, No 1., pp 287 _ 293,2014. SciBulCom Ltd

LINK: http://www.jepe-journal.info/vo115-no-1-2014

OSTALI RADOVI VEZANI ZA REZULTATE ISTRAZIVANJA IZ DOKTORSKE DISERTACIJE:

[2] S.Duli, B.Krstajic "Parallel processing of wind speed data during years 2012 _ 2013 in Shkodra region" in the Scientific Bulletin of Faculty of Natural Sciences, University of Shkodra, volume 64. March 2014

[3] SDuli, B.Krstajic "Parallel computing of Weibull distribution parameters" in the Scientific Bulletin of Faculty of Natural Sciences, University of Shkodra, volume 62. ISSN : 2221-6847, http://80.78.68.65:1000/buletin/01_Bul_62.pdf, March 2012

[4] S.Duli, B.Krstajic, "Hybrid MPI/Pthread parallelization of the Weibull distribution parameters estimator", XIX Scientific-Professional Information Technology Conference, Zabljak, February 2014

[5] S.Duli, B.Krstajic "Methods for estimating the parameters of the Wei bull distribution" in 2nd International Conference "Research and education in natural sciences" Shkodra , ISBN 978-9928-4135-5-0, November 2013

[6] SDuli, B.Krstajic "MPI in a Weibull distribution parameters estimation" International Conference "Towards future sustainable development" Shkodra, ISBN 978-9928- 4135-0-5, November 2012

[7] S.Duli, B.Krstajic, "Pthread u estimaciji parametara weibull distribucije", XVII Scientific-Professional Information Technology Conference, Zabljak, February 2012

[8] S.Duli, B.Krstajic, "Parallel database processmg approaching", International Conference "Challenges of European Economic Integration of Western Balcan" ISBN 978-9928-4011-2-0, Shkoder, Albania, December 2010.

Obrazlozenje mentora 0 koriscenju doktorske disertacije u publikovanim radovima

Radovi iz okvira doktorske disertacije, u kojima je doktorand Sidita Duli prikazala dobar dio ostvarenih rezultata, objavljeni su u: jednom medunarodnorn casopisu sa SCI/SCI-exp liste (1 rad) , jednom casopisu regionalnog karaktera (2 rada) i prezentovani na medunarodnim konferencijama (6 radova). Posebno se izdvajaju 3 rada, od kojih je jedan publikovan u casopisu sa SCI liste, u kojima su predstavljeni kljucni rezultati sprovedenih istrazivanja.

U radu "Parallel Implementation of the Weibull Distribution Parameters Estimator", objavljenom u Casopisu The Journal of Environmental Protection and Ecology (fEPE), ISSN 1311-5065, VoLl5, No 1., pp 287 _ 293, (SCI Impact Factor (2014) = 0.838 / 5-Year Impact Factor = 0.611), izdvojeni su i prezentovani rezultati komparativne analize ubrzanja estimacije parametara Weibull distribucije primjenom dva metoda paralelizacije (MPI i POSIX thread) u procesu modelovanja snage vjetra na odredjenoj lokaciji. Prezentovani su rezulatati istrazivanja i izvedeni zakljucci 0 uticaju kolicine podataka i broja CPU jezgara na ubrzanje procesa estimacije parametara.

Obrazac D2: Ispunjenost uslova doktoranda 2/3

-, UCG

UNlVERZITET eRNE GORE Obrazac D2: Ispunjenost uslova doktoranda

U Podgorici, 25.01.2018. godine

Dobar dio rezultata iz teze, koji se ticu izbora najboljeg metoda paralelizacije koriscenjern seta realnih i zvanicnih podataka 0 brzini vjetra na lokaciji Skadar (Albanija) tokom 2012. i 2013. godine, objavljen je u radu "Parallel processing of wind speed data during years 2012 - 2013 in Shkodra region" in the Scientific Bulletin of Faculty of Natural Sciences, University of Shkodra, volume 64. March 2014. U radu su publikovani rezultati ubrazanja procesa estimacije koriscenjern MPI metoda paralelizacije na realnim podacima za brzinu vjetra u navedenom regionu i periodu. Prezentovani rezultati daju smjernice za koriscenje MPI metode za paralelizaciju u slucaju koriscenja manjeg broja CPU jezgara i potvrduju primjenjljivost ove metode na realni set podataka koji ce se kasnije koristiti i za druge metode i istrazivanja koja su prezentovana u samoj tezi.

Kako su rezultati ukazivali da niti jedan metod paralelizacije ne daje najbolje ubrzanje za razlicite setove podataka i broj koriscenih CPU jezgara distribuiranog racunarskog sistema, dio rezultata istrazivanja koji se tice kombinovanja dva metoda paralelizacije prezentovan je u okviru rada "Hybrid MPI/Pthread parallelization of the Weibull distribution parameters estimator" na 19. Scientific-Professional Information Technology Conference, na Zabljaku u februaru 2014. U ovom radu je prezentovan nacin kombinovanja 2 metoda (MPI i Pthread), rezultatai ubrzanja estimacije, uticaj broja jezgara i kolicine podataka na iste kao i prednosti i ogranicenja hibridnog metoda. Ovi rezultati su polazna osnova za dio istrazivanja u disertaciji koji pokazuje kombinovanje OpenMP i MPI metoda paralelizacija u razmatranoj estimaciji.

Datum i ovjera (pecat i potpis odgovor

Prilog dokumenta sadrii:

1. Potvrdu 0 predaji doktorske disertacije organizacionoj jedinici

2. Odluku 0 imenovanju komisije za pregled i ocjenu doktorske disertacije

3. Kopiju rada publikovanog u casopisu sa odgovarajuce liste

4. Biografiju i bibliografiju kandidata

5. Biografiju i bibliografiju clanova komisije za pregled i ocjenu doktorske disertacije sa potvrdom 0 izboru u odgovarajuce akademsko zvanje i potvrdom da barem jedan Clan komisije nije u radnom odnosu na Univerzitetu Crne Gore

3/3 Obrazac D2: Ispunjenost uslova doktoranda

Univerzitet erne Gore v

ELEKTROTEHNICKI FAKULTET UCG 81 000 ~oclgoricc1, Oz. Vasillgtona bb, tel. (020) 245 839. fax: (020) ~)45 873

Z.R. 510-255-51, PIB: 02016702 302, PDV: 30/31-03951-6 Univerzitet erne Core

Na osnovu sluzbene evidencije i dokumentacije Elektrotehnickog fakulteta u Podgorici, izdaje se

POTVRDA

Mr Sidita Duli, student doktorskih studija na Elektrotehnickom fakultetu u Podgorici, dana 18.12.2017. go dine dostavilaje ovom Fakultetu doktorsku disertaciju pod nazivom: "An approach for the parallelization of the Weibull distribution parameters estimators with applications", na dalji postupak.

Univerzitet erne Gore v

ElEKTROTEHNICKI FAKULTET UCG a 000 Pocj~Jorica, OZ, VasingtOllalJb, tel, (020) 245 839, fax: (020) 245 8"73

z.a 510-255-51, PIS: 02016"702 302, PDV: 30/31-03951-6 Univcrzitet erne Gore

Na osnovu clana 64 Statuta Univerziteta erne Gore, u vezi sa clanom 55 Pravila doktorskih studija, na predlog Komisije za doktorske studije, Vijece Elektrotehnickog fakulteta u Podgorici, na sjednici od 23.01.2018. go dine, donijelo je

ODLUKU

I Utvrduje se da su ispunjeni uslovi iz Pravila doktorskih studija za dalji rad na doktorskoj disertaciji "An approach for the parallelization of the Weibull distribution parameters estimators with applications" kandidatkinje mr Sidite Duli.

II Predlaze se Komisija za ocjenu navedene doktorske disertacije, u sastavu:

1, Dr Milutin Radonjic, vanredni professor Elektrotehnickog fakulteta U niverziteta erne Gore,

2. Dr Bozo Krstajic, redovni profesor Elektrotehnickog fakulteta Univerziteta erne Gore,

3, Dr Slavko Gajin, docent Elektrotehnickog fakulteta Univerziteta u Beogradu,

Komisija iz tacke II ove Odluke podnijccc Izvjestaj Vijecu Fakulteta u roku od 45 dana od dana imenovanja.

-VIJECE ELEKTROTEHNICKOG FAKULTETA-

Dostavljeno: Senatu,

- Centru za doktorske studije, - u dosije, - a/a.

List of articles related to PhD research

Sidita Duli

PhD candidate

Papers in scientific journals

• S.Duli, B.Krstajic "Parallel implementation of the Weibull distribution parameters estimator", in Journal of Environmental Protection and Ecology (JEPE), ISSN 1311-506, March 2014

o Journal Impact Factor (2016) = 0.774

• S.Duli, B.Krstajic "Parallel processing of wind speed data during years 2012 _ 2013 in Shkodra region" in the Scientific Bulletin of Faculty of Natural Sciences, University of Shkodra, volume 64. March 2014

• S.Duli, B.Krstajic "Parallel computing of Weibull distribution parameters" in the Scientific Bulletin of Faculty of Natural Sciences, University of Shkodra, volume 62. ISSN: 2221-6847, http://80.78.68.65:1000/buletinl01_Bul_62.pdf, March 2012

Papers in scientific conferences

• S.Duli, B.Krstajic, "Hybrid MPIIPthread parallelization of the Weibull distribution parameters estimator", XIX Scientific-Professional Information Technology Conference, Zabljak, February 2014

• S.Duli, B.Krstajic "Methods for estimating the parameters of the Weibull distribution" in 2nd International Conference "Research and education in natural sciences", ISBN 978- 9928-4135-5-0, November 2013

• S.Duli, B.Krstajic "MPI in a Weibull distribution parameters estimation" International Conference "Towards future sustainable development", ISBN 978-9928-4135-0-5, November 2012

• S.Duli, B.Krstajic, "Pthread u estimaciji parametara weibull distribucije", XVII Scientific- Professional Information Technology Conference, Zabljak, February 2012

1

Journal of Environmental Protection and Ecology 15, No 1,287--292 (2014)

Computer applications on environmental information system

PARALLEL IMPLEMENTATON OF THE WEIBULL DISTRIBUTION PARAMETERS ESTIMATOR

S. DULIa*, B. KRSTAJIO

=Department of Computer Sciences, Faculty of Natural Sciences, University 0/ Shkoder, Shokdra, Albania E-mail: [email protected] 'Department 0/ Electrical Engineering, University of Montenegro, Podgorica, Montenegro

Abstract. The Weibull distribution is a statistical tool used to model the wind speed. Estimating the Weibull parameters. when the sample of data is given. requires some mathematically complex calcu lations. In case when the sample contains a big amount of data, making these calculations to require longer time. A solution for a better speed up of processing might be to perform these calculations in a parallel computing system. Among different parallelisation strategies, in this research there are implemented two of them: the MPI (message passing interface) and the Pthread. The aim of this article is to verify the increased speed-up ofthe parallel version. and to compare the techniques of paralleling the code, the MPI and the PO SIX, in estimating the Weibull parameters oflarge datasets.

Keywords: parallel programming. the Wcibull distribution, MPL Pthread.

AIMS AND BACKGROUND

The analysis of the wind speed is useful in many fields of industry, in agriculture and in meteorology. Generating energy from the wind power is a mature technol ogy and economically competitive with most fossil fuel applications. Europe is a leader of wind energy using in the world'. The wind energy power is the energy that has been making the fastest increase in the world-. In agriculture, irrigation of an agricultural field above the level of a dam or pond by a pump working by wind energy is much more economical than the pumps working with fuel and electri cal energy", In the field of wind erosion of soil is highlighted that strong winds cause a number of unfavourable conditions and processes in the microclimate and the ecological production environment'. All these highlight the important role of statistics of the wind speed in a specific location.

The Wei bull distribution is a statistical tool used to model the wind speed. The wind speed data collected in a weather station tend to have the form of an Weibull distribution. The shape of the graphics depends on the parameters of the distribution, which are estimated by the set of wind speed data. Once the Weibull

• For correspondence.

287

distribution can be used to calculate the probability of a particular wind speed at a particular location, it can be used to work out the number of hours per year that certain wind speeds are likely to record and therefore the likely total power output of a wind turbine per years.

The aim of this research is to perform the estimation of the Wei bull param eters in a distributed computer system. The principle of parallel programming is to divide the problem in smaller parts and to distribute the calculations between many processors. In this way can be archieved a faster result than at the same code executed in one processor. There are different ways of paralleling the code of this application. Two of them are the MPI and Pthread.

EXPERIMENTAL

Two main methods that can be used to estimate the Wei bull parameters are the maximum likelihood MLE, the least squares, and the Bayesian estimation approach. In this research it is applied the maximum likelihood method, as it is usually con sidered to be more robust and produces more accurate results. MLE is a commonly used procedure because it has very desirable properties", It is used to find global optimum parameters for a certain function to fit a given data set. In this case, MLE is used to find the optimum parameters for the Wei bull distribution.

Among the methods of paralleling the code, are chosen the MPI and Pthreads. MPI provides source-code portability of message-passing programs written in C or FORTRAN across a variety of architectures. A good point of using the message passing is its wide portability. The programs using MPI libraries may run on distributed-memory multicomputer, shared-memory multiprocessors, networks of workstations, and combinations of all of these. MPI is implemented on a great variety of machines, including those 'machines' consisting of collections of other machines, parallel or not, connected by a communication network?

Meanwhile the Pthread library provides an interface to generate and interact with separate threads of execution within a program. This standard is defined by the Institute of Electrical and Electronics Engineers (IEEE) and is available across nearly all variants of the UNIX by IEEE operating systems".

The algorithm is written in C programming language. In the MPI version it is included the mpi.h library and in the Pthread version is included the Pthread.h library. The system where it is executed is a grid cluster, a Linux-based system and portable enough to run with consistent result any implementation of message passing model.

The application takes a censored set of data, which might be ordered or un ordered, taken from a sample of N data. It is given the location parameter of the Wei bull distribution. The data sample and the location parameter are read from a file by each process that is created. The algorithm makes an estimation of the scale parameter and shape parameter of this distribution.

288

Also for the density function of the Weibull distribution is calculated the mode, the mean and the variance for the given set of data. This procedure is repeated R times for different set of data, taken from this N data sample.

In tests performed, the data samples were not taken for a specified location. As the aim of the tests were the to compare the performance and speed-up of these two techniques, and not the wind statistics for any region, the samples were taken for a random wind speed values in-between 1 and 20 km/h. Of course, this implementation could be used for any sample data to any region.

RESULTS AND DISCUSSION

Tests are performed with data samples ofl0 000 000 values and 5 000 000 values. The Weibull parameters are calculated for 100 such samples, taken randomly from a bigger amount of data, from 15 000 000. In the case when the sample contains wind speed data, these samples might be chosen for calculating 100 estimations of 100 different periods oftime, seasons or weeks that might be of interest to research. For each of these samples is calculated also the mode, the mean and the variance of the distribution. The performance is tested for different number of processors, beginning with the serial version with only one processor running and doing the calculations, up to twelve parallel processors doing the calculations. The same tests are done with both techniques, MPI and Pthread. In the Pthread version, there are created parallel threads instead of parallel processes. Both MPI and Pthread scale well up to 12 cores.

Results are compared with the related work in the Stamatakis, parallel pro gramming, implementing the MPI and Pthread in the field ofbioinformatics, and paralleling phylogenetic likelihood functionv".

14000

12000 ~ -+- MPI sample 10000 000 Pthread sample 10 000 000

10000

s 8000

.~ 6000 ., 4000 ." 2000

0 2 3 4 5 6

number of cores Fig. 1. Comparison of time performance ofMPl and Pthreads for a sample containing 10000000 data

Figure 1 shows the time spent to estimate the Weibull parameters of a dataset containing 10 000 000 data. Graphics are built for both versions, MPI and Pthreads, and for a number of cores increasing up to 12 cores. For large dataset, both versions

289

reduce the time spent to estimate the result. Executing the serial program, with a large dataset, it is needed 11 718 s to get the parameters of the Weibull distribution for this sample. There results show how parallel programming helps in reducing this time. Both techniques, MPI and Pthread, get the result 1.96 times faster as soon as a second processor is added. The time is reduced up to ] 220 s when is executed the MPI version in ] 2 cores. Meanwhile time spent for the Pthread ver sion using 12 threads is 1022 s.

14

4

___ MPI sample 10000 000

"._ Ptbread sample 10 000 000 12

10

2

0'---'---'---'-------'-----'--'---'--'---'---'---'---' 2 4 9 10 11 12

number of cores Fig. 2. Comparison of speed-up ofMPI and Pthreads for a sample containing 10000000 data

Figure 2 shows the speed-up graphic for these performance tests. This speed up is the number of times that the parallel implementation is faster than the serial implementation. The tests show that in both cases the speed-up increases in a linear form. According to the Gustavson law II , the linear graphic depends also on an al pha coeficient which is the non-parallelisable fraction of any parallel process. The efficiency shows how well-utilised are the processors in the parallel implementa tion. It is a value between zero and one. A parallel efficiency of one corresponds to ideal, linear speed-up. In this case, the alpha coeficient is small, because the implementation is made with a small fraction of non-pall eli sed.

In a similar comparison between Pthreads and MPI, but in a different case of study, Stamatakis notes that: 'On the larger dataset scalability of OpenMP and Pthreads are similar, while the MPI-based version yields slightly sub-linear speed-ups on the Opteron cluster for 7 and 15 worker processes'P, Similar results are shown in this research. On a larger dataset the Pthread has a speed-up almost linear, showing almost an ideal performance. Meanwhile the performance ofMPI gets down when running the application in 9 or more cores.

The cost of manage of a new process requires more system resources than managing a new thread with shared memory with older ones. Making comparisons in tests for both MPI and Pthread version, for a sample of 10 000 000 data, for smaller number of threads/processors the speed up is nearly the same.

290

The same implementation is executed applying different sample containing 5 000 000 data.

___ MPI sample 10 000 000

Pthread sample 10 000 000

4 5 6 7 9 10 11 12

number of cores Fig. 3. Comparison of time performance ofMPI and Pthreads for a sample containing 5 000 000 data

In Fig. 3 there are presented the grapics oftime spent for estimating the Weibull parameters for a sample containing 5 000 000 data, for both versions, MPI and Pthread. Results show that while executing the serial program, the result is obtained after 5857 s. Again, for a different sample, the parallel program shows its efficency in time, as soon as we add a second processor in the MPI version, or as soon as its created a new thread, in the Pthread version. The minimum time spent, for these tests, is during the execution of the parallel versions in 12 cores. In the MPI case time spent is 712 s, and in Pthread, it is needed only 560 s to have the result.

12 -+- MPI sample 10000 000 Pthread sample 10 000 000 10

6

number of cores

Fig. 4. Comparison of speed-up ofMPI and Pthreads for a sample containing 5 000 000 data

2 4 9 10 11 12

Figure 4 shows the curves of speed-up for both MPJ and Pthread versions, applying the calculations on a sample containing 5 000 000 data. Results show that, with a smaller sample, both speed-ups ofMPI and Pthread are reduced. The speed-up of Pthread is no more almost linear, reaching a value at 10.45 while running 12 parallel threads. The speedup of MPI version is also lower than the speedup shown in Fig. 2.

291

CONCLUSIONS

• This paper introduced the benefits of the parallel programms applied in math ematical and statistical calculations. The serial version is far slower in generating the result compared with the parallel version. That shows that the parallel program presented in this research will be very useful in industry or meteorology, or in every other field where is spent a lot of time analysing wind speed of a specific location.

• For more than 9 threads/ processors, the speed-up of the Pthread version is higher than the speed-up ofMPI version. Also, if we compare in time, and that is the scope, having a faster execution, less time is spent during the execution of the Pthread version. This results for both samples used.

• During the process of building the parallel versions of the application, it is noted that the cost of implementation of the parallel program is the complexity of the code and the knowledge of the programmer about the mpi.h and Pthread.h library. The programmer should care about how the messages will pass without causing any deadlock between sending and receiving new messages. Also the programmer should know the implementation and syncronisation of threads.

REFERENCES 1. K. STEPANOVA: Renewable Energy in Tourism: Opportunities and Benefits. J Environ Prot

EcoL 10 (2), 468 (2009). 2. F. CATOVIC, M. BEHMEN, E. ZLOMUSICA: Trends in the Development of the Electric Power

Systems Based on Wind Energy in World and in Bosnia and Herzegovina. J Environ Prot Ecol, 5 (4), 836 (2004).

3. l. BECENEN, H. INCE K. KARACAVUS: Water Storing for Irrigation of Agricultural Fields from Ponds or Dams by Using Wind Energy in the Thrace Region. J Environ Prot Ecol, 8 (4), 849 (2007)

4. B. PEEY, D. IVANOVA: Climatic and Agroecological Preconditions for Wind Erosion in Bul- garia. J Environ Prot Eeol, 2 (3), 649 (2001).

5. http://www.reuk.co.uk/Wind-Speed-Distribution-Weibull.htm. 6. H. RINNE: The Weibul! Distribution. A Handbook. Chapman & Hall, CRe, 2009. 7. Y. AOYAMA, J. NAKANO: Practical MPI Programming. International Technical Support

Organization, 1999. 8. D. R. BUTENHOF: Programming with POSIX thread. Addison-Wesley, 1997. 9. W. PFEIFFER, A. STAMATAKIS: Hybrid MPIIPthreads Parallelization of the RAxML Phyloge

netics Code. In: IEEE International Symposium on Parallel & Distributed Processing, Workshops and Phd Forum, 2010, 1-8.

10. A. STAMATAKIS, M. OTT: Exploiting Fine-grained Parallelism in the Phylogenetic Likelihood Function with MPI, Pthreads, and OpenMP: A Performance Study. Lecture Notes in Comp Sci, 5265,424 (2008).

11. T. RAUBER, G. RlJNGER: Parallel Programming for Multicorc and Cluster Systems. Springer, 2011.

Received 9 January 2014 Revised 7 February 2014

292

esPARALLEL PROCESSING OF WIND SPEED DATA DURING YEARS 2012-2013 IN SHKODRA REGION

Sidita Duli, Faculty of Natural Sciences, University of Shkoder, Albania Bozo Krstajic, Department of Electrical Engineering, University of Montenegro

Abstract: The Weibull distribution statistical tool is one of different techniques of analyzing the wind speed data. Estimating the Weibull parameters, when the sample of data is given, requires some mathematically complex calculations. Nowadays, most of these estimations are performed in grid systems, where the data can be processed in parallel from different processors. This helps reaching a faster result for the estimation. The aim of this article is to analyze the performance of the parallel implementation in MP! programming, applied on the concrete wind speed data of Shkodra city during the last two weeks.

Abstrakt : Shperndarja statistikore Weibull eshte nje nga disa teknikat qe perdoren per te analizuar te dhenat e shpejtesise se eres. Vleresimi i parametrave te Weibull, per nje master te dhene, kerkon perllogaritje komplekse matematikore. Ne ditet e sotme, shumica e ketyre vleresimeve realizohen ne sisteme grid, ne te cilet te dhenat mund le perpunohen ne disa procesore njeheresh. Kjo ndihmon ne marrjen e nje rezultati me te shpejte. Qellimi i keti] artikulli eshte analizimi i performances te implementimit ne programimin MP!, i aplikuar per te dhena konkrete te shpejtesise se eres gjate dy viteve te fundit.

INTRODUCTION The analysis of the wind speed is useful in many fields of industry, in agriculture and in meteorology. Generating

energy from the wind power is a mature technology. This is the main reason that highlights the important role of statistics of the wind speed in a specific location.

The Weibull distribution is a statistical tool used to model the wind speed. The wind speed data collected in a weather station tend to have the form of an Weibull distribution. The shape of the graphics depends on the parameters of the distribution, which are estimated by the set of wind speed data. Once the Weibull distribution can be used to calculate the probability of a particular wind speed at a particular location, it can be used to work out the number of hours per year that certain wind speeds are likely to record and therefore the likely total power output of a wind turbine per year [1]. The Weibull distribution exists in two main forms: the two-parameter and three-parameter Weibull distribution. But the study is focused in the two parameters form, as it is the form applied in meteorology. The two parameter Wei bull Distribution has the following density and distribution functions:

(1)

( XJb F(x) = l-e a (2)

The parameter is the Weibull scale parameter in mls; a measure for the characteristic wind speed of the distribution. This parameter is proportional to the mean wind speed. The b is the Wei bull shape parameter. It specifies the shape of a Weibull distribution and takes on a value of between 1 and 3. A small value for b signifies very variable winds, while constant winds are characterized by a larger b [2].

2 METHODS

The application makes an estimation of the parameters of the Weibull distribution function. Two main methods that can be used in this estimation are the maximum likelihood and the least squares. In the implementation used in this study it is applied the maximum likelihood method, as it is usually considered to be more robust and produces more accurate results. The computer application that estimates the Weibull parameters takes too much time to produce a result, especially when the sample contains too much data. This is one reason to use the grid architecture and to process the data in multiple cores simultaneously. The principle of parallel programming is to divide the problem in smaller pieces and to distribute the calculations between many processors. Using the technique of parallelizing the code by two different processors that use the same memory space, it makes possible that complex tasks might be computed in a shorter time as they were computed by a single processor architecture. . . There are different ways of para lie liz at ion the code of this application. Two of them are the message passing mterface (MPI) and Posix threads (Pthread). In this research it is analyzed the MPI implementation of the parameter estimation of Wei bull parameters.

MPI is a standard developed by the Message Passing Interface Forum (MPIF). It specifies a portable interface for writing message-passing programs, and aims at practicality, efficiency, and flexibility at the same time. MPI is implemented on a great variety of machines, including those "machines" consisting of collections of other machines, parallel or not, connected by a communication network [3]. The programs using MPI libraries may run on distributed memory multicomputer, shared-memory multiprocessors, networks of workstations, and combinations of all of these. The algorithm is implemented in the C programming language. It is imported the mpi.h library which contains the parallel procedures and function of the communication and passing the messages between cores of the system. The system where it is executed is a grid cluster, a Linux based system and portable enough to run with consistent result any implementation of message passing model. The application takes a censored set of data, which might be ordered or unordered, taken from a sample of N data. The data analyzed in this research are the wind speed data collected from the weather station of the team of learning technique of University "Luigj Gurakuqi" for a two year period 2012-13. It is given the location parameter of the Weibull distribution. The data sample and the location parameter are read from a file by each process that is created. The algorithm makes an estimation of the scale parameter and shape parameter of this distribution.

4.RESULTS AND DISCUSSIONS It is of interest to study the performance of this implementation in parallel of Weibull parameter estimation for a concrete wind speed data in our location. The weather station is installed in the city of Shkodra. It measures the wind speed every 30 minutes. Also it is of interest to analyze the distribution of the wind speed of the last two years. For this reason the parallel implementation is applied to three special data samples. The first data sample is the wind speed data of the 2012. This sample contains 17568 data. The second sample is the wind speed data during the year 2013, and contains 17520 data. The third sample is that containing both wind speed data of years 2012 and 2013. It contains 35088 data. The samples in these tests are not the best to show the parallel programming advantage of decreasing the time of calculations because the samples are not containing too much data to be complex calculations. One reason of performing the tests on these samples is because of the meaning of the data, as they are concrete wind speed data of a particular location, and not random data. This is a first step to a concrete application of this parallel version. The MPI parallel implementation is tested for different number of cores, from two up to twelve cores. The results are compared to the serial version represented one cored version in the above tables.

a) Tests of performance when the data sample contains wind speed data of the year 2012 in Shkoder

Cores 1 2 3 4 5 6 7 8 9 10 11 12 Time in seconds 35.7 19.1 13.6 9.8 7.7 6.3 5.4 4.8 4.3 3.9 3.6 3.5 Speed-up 1 1.86 2.62 3.64 4.63 5.66 6.61 7.43 8.3 9.15 9.91 10.02

Table I: Performance in time and speed up of the MPI implementation proccessmg wmd speed data dunng 2012.

Figure 1 : Speed up of the MPI version for the sample wind speed data of the year 2012 Results show that the time spent to perform the calculations is reduced while increasing the number of cores.

For this estimation the serial version is executed in 35.7 seconds. The tablel shows the time spent to estimate the Weibull parameter; of a dataset containing 17568 data. For this dataset, it is shown that the MPI version reduce~ the time spent to estimate the result. Executing the serial program it is needed _35.7 s to ~et the p~amet~rs ?fthe Weibull distribution for this sample. There results show how parallel programmmg helps in reducmg this time, The MPI technique the result 1.86 times faster as soon as a second processor is added. The time is reduced up to only 3.5 s when is executed the MPI version in 12 cores.

This table shows the speed-up graphic for these performance tests. This speed up i~ the ~umber of times that the parallel implementation is faster than the serial implementation. The tests show that m this case the speed up

14

12

10

c. 8 .l-- :::l "C G1 G1

6 Co T

'" 4

2

0 -,- 1 2 3 4 5 6 7 8 9 10 11 12

- MPI version speed up ~_-------~~~-----------

-Linear_speed \-IP

Number of cores

increases in a linear form. According to the Gustavson law f4], the linear graphic depends also by an alpha coeficient which is the non-parallelisable fraction of any parallel process. The efficiency shows how well-utilized are the processors in the parallel implementation. It is a value between zero and one. A parallel efficiency of one corresponds to ideal, linear speedup. In this case, the alpha coeficient is small, because the implementation is made with a small fraction of non-pallelizied.

Results of parameter estimation of Weibull distribution for this sample show that the shape parameter is b=3.358281 and the scale parameter is a=2.628137.

b) Tests of performance when the data sample contains wind speed data of the year 2013 in Shkoder

Cores 1 2 3 4 5 6 7 8 9 10 11 12 Time in seconds 34.9 18.6 l3.3 9.3 7.3 6.1 5.l 4.6 4.2 3.9 3.6 3.4 Speed-up 1 l.87 2.62 3.75 4.78 5.72 6.84 7.58 8.31 8.94 9.69 10.26

Table 2: Performance m time and speed up of the MPI Implementation proccessmg wind speed data dunng 2013.

Figure 2: Speed up of the MPI version for the sample wind speed data of the year 2013 For this estimation, the serial version is executed in 34.9 seconds. The table 2 shows the time spent to estimate the Weibull parameters of a dataset containing 17520 data. The serial implementation needs 34.9 s to get the parameters of the Weibull distribution for this sample. Time is reduces as soon as the second procces is added. When the number of cores is two, it takes 18.6 s to perform the calculations, and the speedup is 1.87. As the number of cores is increased, it reaches the best performance in the seventh core, as it is almost linear speedup.


14

12 ----------------------------------------------~~

____ -_ Linear speed l~ _

o ~--~------~------~------~--~--.-------~--- 7 8 9 10 11 12 5 6 2 3 4 1

Number of cores

c) Tests of performance when the data sample contains wind speed data of both years 2012 and 2013 in Shkoder

Cores 1 2 3 4 5 6 7 8 9 10 11 12 Time in seconds 66.7 34.3 23.3 17.2 14.7 12.6 10.5 9.4 8.1 7.3 6.6 6.4 Speed-up 1 1.94 2.86 3.87 4.53 5.27 6.35 7.09 8.23 9.11 10.1 10.4

Table 3: Performance in time and speed up of the MPI implementation proccessmg wind speed data dunng both years 2012 and 2013.

3

2 3 4 5 6 7 8 9 10 11 12

14

12

10

Q. 8 :s " QJ QJ 6 _,__ Q. II>

4

2

0 1

-Linear speed up - --

Number of cores Figure 3 : Speed up of the MPI version for the sample wind speed data of the year 2012 and 2013

For this estimation, the serial version is executed in 66.7 seconds. The table3 shows the time spent to calculate the Weibull parameters of a dataset containing 35088 data. The serial program needs 66.7 s to get the parameters of the Weibull distribution for this sample. The time is reduced up to only 6.4 s when is executed the MPI version in 12 cores. The figure 3 shows that the speed up decreases when the number of cores is larger, but still it tends to be close to the linear speed up.


5.CONCLUSIONS The paper introduces the benefits of implementing the parallel processing in analysing concrete data in meteorology. In this case the Weibull distribution parameters are estimated to represent the distribution of the wind speed data for two last years. It is tested the performance in time of the MPI parallel version. Also it is measured the speed up of the execution, from two up to twelve cores. Time of the estimation is reduced as soon as adding the second core. When the application is run in twelve parallel processes, time is only few seconds. The cost of adding new procceses, instead of using only one process in the serial version, is the complexity of the code, and the effort of the programmer to make it work by using the message passing. Adding new procceses slows the execution as a part of this time takes the communication between procceses. As future work might be testing the parallel implementation with wider range of concrete data, which might be including more years in analyze, or comparing different locations. In this way, having wider range of data, it is highlighed the importance of introducing the parallel programming in the statistical estimations.

6. REFERENCES 1. http://www.reuk.co.uklWind-Speed-Distribution-Weibull.htm. 2. The Swiss wind power data website bttp:llwind-data.ch/ 3. Y.AOYAMA, 1. NAKANO: Practical MPI Programming, International Technical Support

Organization, 1999 4. T. RAUBER, G. RUNGER: Parallel Programmingfor Multicore and Cluster Systems. Springer, 2011.

HYB RID MPIIPTHREAD P ARALLELIZATION OF THE WEm ULL DISTRm UTION PARAMETERS ESTIMATOR

S.Duli a, B.Krstajic b a Department of Computer Sciences, Faculty of Natural Sciences, University ofShkoder, Albania

b Department of Electrical Engineering, University of Montenegro, Podgorica, Montenegro Email: [email protected]. [email protected]

Abstract: In this paper it is presented the hybrid programming implementation ofMPI and Pthreads to estimate the parameters of Weibull distribution for a given data sample. Message passing communication is managed by MPl to achieved coarse grained parallelism between nodes, and shared memory communication is implemented by Pthreads. The aim of this

article is to verify the speedup of the hybrid MPUPthread version, and to compare it to the techniques ofparallelizing the code, the MPI and the POSlX, in estimating the Weibull parameters oflarge datasets.

Keywords: parallel programming, Weibull distribution, MPI, Pthread

Introduction Weibull Distribution is a statistical tool used to model the wind speed. The wind speed data collected in a weather station tend to have the form of a Weibull distribution. The shape of the graphics depends on the parameters of the distribution, which are estimated by the set of wind speed data. Once the Weibull distribution can be used to calculate the probability of a particular wind speed at a particular location, it can be used to work out the number of hours per year that certain wind speeds are likely to record and therefore the likely total power output of a wind turbine per year [1]. The principle of parallel programming is to divide the problem in smaller pieces and to distnbute the calculations between many processors. 1n this way can be achieved a faster result than the same code executed in one processor. There are different ways of parallelization the code of this application. Two of them are the MPI and Pthread. The aim of this research is to represent a hybrid implerrentation of the pararreter es timation of Wei bull parameters. Methods Two methods used in implerrenting the parallel version of the estimation of Weibull parameters are MPI and Pthreads. The programs using MPI libraries may run on distributed rremory multicomputer, shared-rremory multiprocessors, networks of workstations, and corroinations of all of these. MPI is implerrented on a great variety of machines, including those "machines" consisting of collections of other machines, parallel or not, connected by a communication network [2] . Meanwhile the Pthread library provides an interface to generate and interact with separate threads of execution within a program This standard is defined by the IEEE and is available across nearly all variants of the UNIX operating systems [3] .The hybrid version MPIIPthread is implerrented by creating two threads for each process involved in the calculations. The application takes a censored set of data, which might be ordered or unordered, taken from a sample of N data. It is given the location pararreter of the Weibull dis tribution. The data sample and the location parameter are read from a file by each process that is created. The algorithm makes an estimation of the scale parameter and shape parameter of this dis tribution. Also for the density function of Weibull distribution is calculated the mode, the mean and the variance for the given set of data. This procedure is repeated R tines for different set of data, taken from this N data sample.

Results Tests are performed with data samples of 10 000 000 values and 5 000000 values. The Weibull pararreters are calculated for 100 such samples, taken randomly from a bigger amount of data, from 15 000000. For each of these samples, there are calculated the mode, the rrean and the variance of the distribution. The performance is tested for different number of processors, beginning with the serial version with only one processor running a doing the calculations, up to twelve parallel processors doing the calculations. Tests are performed for each implerrentation, Pthread, MPI and Hybrid version with two threads.

Noof Hybrid 2 Pthread Cores threads MPI version Version

2 6421 6022 5960 4 3209 3098 2947 6 2195 2048 1998 8 1632 1567 1502

10 1315 1292 1213 12 1294 1220 1022

Table 1: Time of execution of different version for the sample containing 10 000 000 data. Table 1 shows the tine spent to estimate the Weibull parameters of a dataset containing 10000 000 data. Tests are made for hybrid two Pthreads and one process, up to two Pthreads and 6 processes. For large dataset, both versions reduce the tirre spent to estimate the result. Results show that the tine spent to perform the calculations is reduced while increasing the nurroer of cores. For this estimation, hybrid version executed with two threads and one process takes 6421 seconds. For the same dataset, this version takes 1294 second when executed with two threads and six processes. These results are compared with those of MPI version executed with the sarre number of processes as cores, and those of Pthread version executed with the sarre number of threads as cores. As it is shown in the table 1, for this range of dataset, the Pthread version needs less tine to perform the estimation. Pthread version takes 5960 seconds to estimate the parameters while it is executed with two threads, and 1022 seconds while it is executed with twelve threads. From table 1 we can conclude that MPI pure version nor hybrid MPIIPthread is not better than Pthreads in terms of execution tine] 4].

.,

11.00 0.80

Hybrid 2-threads efficiency

-- 0.60

0.40

0.20

0.00 2 4 6 8 10 12

Nrofcores

Fig 1. Efficiency hybrid MPIIPthread version when data sample contains 10 000 000 data. The figure 2 represents the efficiency of the hybrid MPIIPthread version. Results show that the efficiency is relatively high when the nunoer of cores s small. Adding new processes after the five process decreases the efficiency. Increasing the nurroer of processes, the connnunication overhead becomes outstanding and the parallel performance degrades. The hybrid MPIIPthreads programming IOOde could obtain a good parallel performance when the processes' number is larger by reducing the communication between processes. The sam: implementation is executed applying different sample containing 5 000 000 data.

Noof Hybrid 2 MPI Pthread Cores threads version Version

2 3249 3078 2938 4 1618 1585 1471 6 1137 1120 997 8 819 806 741

10 765 752 625 12 729 712 560

Table 2: Tim: of execution of different parallel versions for a sample with 5 000 000 data. Results show that hybrid version executed with two threads and one process takes 3249 seconds for estimating the Weibull parameters of a sample containing 5 000 000 data. For the same dataset, this version takes 729 second when executed with two threads and six processes. Results are compared with those of MPI version executed with the sane number of processes as cores, and those of Pthread version executed with the sam: number of threads as cores. For this range of dataset, the Pthread version performs faster the calculations. Pthread version takes 2938 seconds to estimate the parameters while it is executed with two threads, and 560 seconds while it is executed with twelve threads.

Hybrid 2-threads efficiency 1.00

0.80 +--------- .....•.... ~"'"'_............__::----- 0.60 -t----------------- 0.40

0.20

0.00 2 4 6 8 10 12

Nr of cores

Fig 2. Efficiency hybrid MPIIPthread version when data sample contains 5000000 data. Figure 2 shows the graphical presentation of the efficiency of the hybrid 2 threads MPIIPthreads version, Results show that for a small nurmer of processes involved in the parallel calculations the efficiency is high. With increasing the number of processes the parallel performance begins to decrease. The graphic shows that the efficiency is still high when there are involved four processes. They are all combined with 2 threads each, distributed in eight cores, give a good performance. Both figure 1 and figure 2 show that for a large dataset the efficiency stays high for bigger nurmer of cores, respectively ten cores. For a smaller dataset, in figure 2, the efficiency is high only until the number of eight cores.

CONCLUSIONS • The hybrid 2 threads MPIIPthread might be as a technique of parallelization of estimation of Weibull distribution. The hybrid implementation performs the calculations in less tine, as the number of processes involved is increased. • The hybrid 2 threads MPIIPthread needs IOOre time to perform calculations than the pure MPI version. Also the hybrid is slower than the pure Pthread version. This s due to communication needed between processes. • The efficiency of the hybrid 2 threads MPIIPthread falls when the number of cores is higher than eight, for a sample 5 000 000. For the sample containing 10 000 000 data, the efficiency falls when the number of cores is higher than ten. • As future work is analyzing the hybrid 4 threads MPIIPthreads and comparing the results with this research.

REFERENCF.S [1 ].http://www.reu k.co.uk/W ind -Speed -Dis tribution Weibull.htm [2].Y.AOYAMA, J. NAKANO

International Practical MPI

Technical Support Programming, Organization, 1999 [3].D. R BUTENHOF: Programming with Pthread, 1997 [4] Huailiang Xuan, Yue Hu, Weiqin Tong, Zhixun Gong and Yan Hou, "Hybrid programming implementation of Rice method on SMP cluster architecture", IEEFJACIS 11 International Conference on Computer and Information Science, 2012

PARALLEL COMPUTING OF WEIBULL DISTRIBUTION PARAMETERS

Sidita Duli, Faculty of Natural Sciences, Uni wrsity of Shkoder, AI bani a Bozo Krstajie, Department of FJectricai Engineering, University of Montenegro

Abstract: The statistical calculations might have complex formulas to compute. In cases when the sample contains a big amount of data, the time of execution increases. One of these is calculating the parameters of Wei bull distribution when the sample data is given. The parameters are estimated by using the maximum likelihood method, which requires many mathematical computations. For a better speed of processing the result, a solution might be to make this calculation in a parallel computing system, and by using a parallel algorithm. One technique is by using the Pthread library in C application. The threads created are used to compute simultaneously the calculations in complex statistical formulas. This article shows how the Pthread code added in the application helps to perform faster.

1. INTRODUCTION The weibull distribution is an important distribution especially in maintainability analysis and reliability. Furthermore it is widely used in biorredical engineering and to model variations in wind speed. This distribution exists in two main fo rms: the two-parameter and three-parameter W eibu 11 dis tribution. The two pararreter Wejbull Distribution has the following density and distribution functions:

(b) o': (X)b f(x) = -;; -;; e--;; (1 )

(2)

where a is the scale paraneter and b is the shape parameter. The three parameter Weibull distribution has the following formula of density and distribution:

f(x) = (£)(-c) b-1e-e:C)b a a (3)

X-c b F(x) = 1- e-C-;;-) (4)

where a is the scale paraneter, b is the shape parameter and c is the location parameter, The application makes an estimation of the parameters of the Weibull distribution function. Two main rrethods that can be used are the maximum likelihood and the least squares. In the application it is used the maximum likelihood method, as it is usually considered to be more robust and produces more accurate results. The method of maximum likelihood [2] is a commonly used procedure because it has very desirable properties. Let xi , X2 , ••• , Xn be a random sample of size n drawn from a probability density function fx (x;e) where e is an unknown parameter. The likelihood function of this random sample is the joint density of the n random variables and is a function ofthe unknown parameter. Thus

L=TIf=l fxi(xi,8) (5) is the likelihood function. The maximum likelihood estimator (MLE) of e , say Ae, is the value of e that maximizes L or, equivalently, the logarithm of L. Often, but not always, the MLE ofe is a solution of

dlogL =0 (6) dO

where solutions that are not functions ofthe sample values xi, X2, ... , Xn are not admissible, nor are solutions which are not in the parameter space. The problem with the MLE lies in obtaining initial estimates for x and e. MLEdoes not supply the values for x and e, those values are supplied by a parameter estimation technique. The most common technique for pararreter estimation of Wei bull distributed reliability data is the Newton-Raphson algorithm (NRA). The Newton-Raphson algorithm uses a Taylor Series expansion about the predefined function to estimate the parameters' values. However, the main problem with the NRA algorithm is the selection process for the initial parameter values. The selection of the initial parameter values is critical, because of the inability of the NRA to avoid convergence to local optima. f(x; e) is the function which is trying to be optimized. In this research the function f(x; e), is the Two-Parameter Weibull Distribution probability density function. These calculations in the serial version requires lots of time when the data sample is too large and when the calculations are done for a large number of samples. The parallel version executed in parallel computers systems

makes these estimations in less time. There are different ways of parallelization the code of this application, but below is presented an applied method by using the POSIX library.

2 METHODS

Using the technique of paralielizing the code by two different processors that use the sane rrermry space. it makes possible that complex tasks might be computed in a shorter tine as they were computed by a single processor architecture. Pthreads are a set of types and procedure calls in C. They are implemented in a pthread.h file, and a thread library. [1] The POSIX thread library provides an interface to generate and interact with separate threads of execution within a program. This standard is defined by the IEEE and is available across nearly all variants of the UNIX operating systems. The processor creates threads that are used to compute a part of calculations of the whole application. They are running concurrently. so the total time to compute the result is smaller and is depending on the number of threaded created by the processor. The primary motivation for using threads is to realize potential program performance gains. Comparing to processes, creating a thread requires fewer operations, and to manage a thread requires fewer system resources. Another advantage is related to the software portability. Applications that use threads can be developed on serial machines and run on parallel machines without changing anything. This portability is very significant advantage of threaded APIs. The program takes a censored set of data, which might be ordered or unordered, taken from a sample of N data. The location parameter A is given, and the algorithm makes an estimation of the scale parameter B and shape parameter C of the equation (4) of the Weibull distribution. The method used for this estimation is the maximum likelihood estinIation. To calculate the root of (6) where the L is the likelihood function and is computed via a Newton-Raphson iteration. The Newton-Raphson method in one variable is implemented as follows: Given a function f defined over the reals x and its derivative f " we begin with a first guess Xo for a root of the function f [3]. The function is reasonably well-behaved a better approximation Xl is

(7)

In our case, the value of Xo is given, and if it is zero is it taken a guess value ofx. Also for the density function is calculated the mode, the mean and the variance for the given set of data. This procedure is repeated for R tines for different set of data, taken from this N data sample. The algorithm is written in C, including the Pthread library. The system where it is executed is a grid cluster, a Linux based system and portable enough to run with consistent result on any POSIX system The data sample and the location parameter is read from a file by each thread that is created. These parallel threads take the set of data M, fro m the s amp le N, and it is repeated Rlnr_ threads tines.

4.RESULTS AND DISCUSSIONS Tests are performed with data samples of 1000000 values, and 2000000 values. The Weibull parameters are calculated for 100 such samples, and for each of them is calculated also the mode, the mean and the variance of the distribution. The project is tested for different number of threads, beggining with the serial version with only one thread running a doing the calculations, to eight threads in parallel doing the calculations. In the table land table 2 are given the results for the time in seconds and for the speed-up for different number of threads used. Table 1: Performance in time and speed up of the application with a sample of 1000 000 data.

Timein Threads Seconds Speed-up

1 1214 1

2 420 2.8

3 269 4.5

4 204 5.9

5 165 7.3

6 135 8.9

7 111 10.9

8 97 12.5

2

Table 2: Perfonnance in time and speed up or the application with a sample of 2000 000 data.


1 2436 1

2 816 2.9

3 537 4.5

4 408 5.9

5 325 7.4

6 260 9.3

7 225 10.8

8 187 13.0

The figure 3 and figure 4 show a graphical presentations for the time spent for different number of threads. Time in reduced imrrediately as the second thread is added, increasing in this way the performance of all the parallel executions of the application.

Figure 3: Graphical presentation of the perfonnance in time of sample with 1000000 data.

1400

1200

..•• 1000 c: § 800 I .E 600 • ! 400

200

0

0 4 6 8 10 2 Number of threads

Figure 4 : Graphical presentation of the performance in time of sample with 2000000 data.

12000 f-~~-f~~~~~~~i-~~~~~~ !1500 ~~~~~~~~~~~~~~±=~~~ • !1000 +-~~~~~~~--~~~~~~~7=~

4 6 8 10

Numberofthreads

The figure 5 shows the speed-up graphic for these performance tests. The tests shows that in both cases the speed up increases in a linear form.

Figure S:Graphical presentation of the perfonnance of speed up

10

:- 8 ~---'-l----'-l--"';'__':-_I-'- --+--,-+--c--'+~z!."" '-l--""'-'-7"-H----+-<-t-++--'-I 11 ! 6 T+-=~r-~~~~~~-~~~-~~~~~~~~~--,---~

o 4

Numberofthreads __ Sample wiIh 2 000 000 dlltzt

••••• sampl. with lOOO000dal4

8 9

o

-.;

5.CONCLlNIONS

This paper introduced the benefits of using the Pthread library as a function of highlighting the advantaged of the parallel versions in mathematical and statistical calculations. The Pthread has the ability to create concurrent excecutions instances on a single processor. It also allows to share global memory and maximizes the machine's resources. In the performance tests it is shown that adding a second thread iy helps a lot by increasing the speed up till 6 times faster. Also it is noticed that after a certain number of thread, the speed up is not increased so fast as it does for a couple threads added. Also it is worth noting the difficulty that is required to tum the serial code into a parallel code using the threads. The programmer has to make this explicitly in the code, by using the Pthreads library functions of creating and managing them It is required more work to the programmer to know how the threads work and how they will share the rnemory between them The programmer should know how to manage the access of a shared variable in a simultaneously way. Also the programrner should be careful with the grobal variables. Threads should not make changes to them, but only by using the mute x semaphore. The cost of creating threads, instead of using only one thread in the serial version, is the complexity of the code, and the effort of the programmer to make it work without deadlocks. However, compared to the cost of creating and managing a process, a thread can be created with much less operating system overhead. Managing threads requires fewer system resources than managing processes.

REFERENCES

[1] B. Lewis, DJ. Berg, (2008), Pthreads Primer.

[2] Harter, H. L and Moore, A. H. (1965). Point and Interval Estimation, Based on m-order Statistics, for the scale pararneter ofa Wiebull Population with Known Shape parameter. [3] http://en.wikipedia.org/wikiiNewton%27s_rnethod

A

METHODS FOR ESTIMATING THE PARAMETERS OF THE WEffiULL DISTRffiUTION

Sidita Duli, Bozo Krstajic

Faculty of Natural Sciences, University of Shkoder, Albania Department of Electrical Engineering, University of Montenegro

[email protected] [email protected]

ABSTRACT

The Weibull distribution is used in meteorology as representing the wind speed frequency distribution. There are dtfferent methods for estimating the two Weibull parameters ( scale parameter c and shape parameter k)from wind statistics. The algorythm of these methods can be implemented in a computer by using a programming language so it can estimate the Weibull parameters for dtfferent givven samples. This article presents the Weibull main methods for estimating the nw Weibull parameters. It also shows the performance of a C application that makes the calculations.

INTRODUCTION

Weibull Distribution is highly used in meteorology to model the wind speed. This statistical tool tells us how often winds of different speeds will be seen at a location with a certain average wind speed. This distribution is used also wind energy applications. In this article are shown the reasons why is this distribution used for this purpose and how can the parameters estimated. Based on the mathematical calculations, it is build and tested a serial computer application that estimates these parameters. The variation in wind speed are best described by the Weibull probability distribution function 'h' with two parameters, the shape parameter 'k', and the scale parameter 'c'. The probability of wind speed being v during any time interval is given by the following:

( )( )k-l (v)k

h(v) = ~ ~ e- ~ for 0 <v < 00 (1)

In the probability distribution chart, h is plotter against v over a chosen time period, where :

h=fraction of time wind speed is between vand (v + Av) / ~v

is the plot of h versus v for three different values of k. The curve on the left with k = 1 has a heavy bias to the left, where most days are windless (v=O). The curve on the right with k = 3 looks more like a normal bell shape distribution, where some days have high wind and equal number of days have low wind. The curve in the middle with k = 2 is a typical wind distribution found at most sites. In this distribution, more days have lower than the mean speed, while few days have high wind. The value of k determines the shape of the curve, hence is called the 'shape parameter'. The Weibull distribution with k = I is called the exponential distribution which is generally used in the reliability studies. For k>3, it approaches the normal distribution, often called the Gaussian or the bell-shape distribution. For greater values of c, the curves shift right to the higher wind speeds. That is, the higher the c, the more number of days have high winds. Since this shifts the distribution of hours at a higher speed scale, the c is called the scale parameter. (C.G.Justus, W.R. Hardgraves, A.Mikhail, D.Graber, 1977)

Advantages of the Weibull distribution are that: 1. it is a two-parameter distribution, depending only in c and k, 2. in a wide number of cases the Weibull seems to give a reasonable fit to observed distributions. 3-

with Weibull c and k values known at one height, a consistent methodology can be used to adjust these paramters to another desired height.

METHODS

There are several methods which can be used to estimate the Weibull parameters c and k, depending on which wind statistics are available and what level of sophistication in data analysis one wishes to employ. (Rinne, Horst,2009)

Least -squares

Least square method is used to calculate the parameters in a formula when modeling an experiment of a phenomenon and it can give an estimation of the parameters. When using least square method, the sum of the squares of the deviations s which is defmed as below, should be minimized

11

S = L w;[y; - g(x;)Y i~l

(2)

in the equation, xi is the wind speed, yi is the probability of the wind speed rank, so (xi.yi) mean the data plot, wi is a weight value of the plot and n is a number of the data plot. The estimation technique we shall discuss is known as Linear Least Square Method (LLSM) which ia a computational approach to fitting a mathematical or statistical model to data. Ii is applied in engineering and mathematics problem that is often not thought of as an estimation problem. The linear least square method (LLMS) is a special case for the least square method with a formula which consists of some linear functions and it is easy to use. An in the more special case that the formula is line, the linear least square method is much easier.

Maximum likelihood

The method of maximum likelihood [2] is a commonly used procedure because it has very desirable properties. Let xi , X2 , ••• , x, be a random sample of size n drawn from a probability density function fx (x;e) where e is an unknown parameter. The likelihood function of this random sample is the joint density of the n random variables and is a function of the unknown parameter. Thus

(3)

is the likelihood function. The maximum likelihood estimator (MLE) of e, say 8, is the value of e that maximizes L or, equivalently, the logarithm of L . Often, but not always, the MLE of e is a solution of

dlogL =0 dO

(4)

where solutions that are not functions of the sample values xi, X2, .•• , X" are not admissible, nor are solutions which are not in the parameter space. Applying the MLE to estimate the Weibull parameters, namely the shape parameter and the scale parameter. Consider the Weibull probability density function given in (1) then likelihood function will be

n (kJ(X. )k-l -( Xi r L(x)'x2, ••• ,xn,k,c)= n - _I e C (5) 1=1 C C

on taking the logarithms of (5), differentiating with respect to k and c, in turn and equating to zero, we obtain the estimating equations

8lnL n" 1 11

--=-+ I)nx: -_ Lxklnx. =0 8k k ;=1 1 C i=1 I I

(6)

aln L - n 1 11 k --=-+-"x. =0 a 2L..." ccc ;=1

(7)

On eliminating c between these two above equations and simplifying, we get

II "x~ ln x, £..... I t 1 1 11

i=1 ---_ "lnxt• =0 11 k £..... LX; n i=1 i=1

(8)

which may be solved to get the estimate of k. This can be accomplished by Newton-Raphson method, which can be written in the form:

x =X _ !(xn) • n-I 11 !'(xn) (9)

where

1l "x~ ln ,r. £..... I 1 1 1 11

!(k)=i=l" ----Llnxi "k k n ;=1 z:».

(10)

;=1

and

(11)

once k is determined, ca can be estimated using equation (7) as

11 LX; c = £!.__

n (12)

There is build an C program, a serial version, that based on a sample written in a text file, are calculated the two parameters of Weibull distribution. The program takes a censored set of data, which might be ordered or unordered, taken from a sample of N data. The location parameter a is given, and the algorithm makes an estimation of the scale parameter c and shape parameter k of the equation (1) of the Weibull distribution. The method used for this estimation is the maximum likelihood estimation. To calculate the root of (4) where the L is the likelihood function and is computed via a Newton-Raphson.

RESUL TS AND DISCUSSION

Different tests are made to measure the time needed to calculate the parameters for different data samples. There are used two types of data : random data and concrete data. Random data samples are wide in dimension, containing over 1 million data, all random numbers. The purpose of taking these samples is testing how long in time can perform the execution of a wide data sample. Tests are performed with two

samples, one containing 10 million data, and another containing 5 million data. Two other concrete data samples are tested. One data sample contains concrete numeric values of wind speed measured in a weather station in Podgorica, Montenegro, This sample contains 105122 data. These are values are collected from the weather station during the whole years 2010-2011, measuring the wind speed every 10 minutes and saving these data in a text file. Another data sample contains values of the wind speed measured in the weather station located in University of Shkodra, Albania. This sample contains 17088 data. This station measures the wind speed every 30 minutes, and saves the results in a text file too. The data used from this station are for the period from 1 June 2010 to 20 May 2011.

Table l:performance in time for the serial version for four tested samples ----_----

Testing the performance of this application results that with increasing the size of the sample, increases the time of calculating the Weibull parameters for the given sample. The time varies from 171 seconds when the sample is 17088 data, up to 11718 seconds for a wide sample containing 10 million data. Also tests shows that there is no difference if the sample contains random data or concrete data.

Performance in time for serial version 14000

11718 - ~

- 5857 ~

~ I 1502 I 171

This paper introduces the methods used for the Weibull distribution estimating parameters. Based on one of them, the maximum likelihood method, is built an application in C that calculates the parameters for a given sample. Different tests are made to calculate the performance of this application. Results show that the time needed to get the result depends on the size of the sample. When the sample is at a very large size, it takes too long to have the result. In this case, the performance of the serial version is not at all optimal. For small samples, the application can have a faster result, which can vary to hundreds of seconds. In order to get faster the result, may be the implementation of the algorithm in a parallel architecture.

12000

III 10000 -c c 8 8000 ~ .~ 6000 E 1= 4000

2000

o 10000000 5000000 105120 17088

Methods for estimating wind speed frequency distributions, e.G.Justus, W.R. Hardgraves, A. Mikhail , D.Graber, Journal of applied meteorology, 1977

The Weibull distribution, a handbook, Rinne, Horst. Chapman & HaWCRC,2009

Size of the data sample

CONCLUSIONS

REFERENCES

PfHREAD U ESTIMACIJI PARAMETARA WEIBULL DISTRIBUCIJE

PfHREAD IN A WEffiULL DISTRffiUTION PARAMETERS ESTIMATOR

Sidita Duli, Faculty of Natural Sciences, Uni verstty of Shkoder, Albania Bozo Krstajic, Department of Electrical Engineering, Universtty of Montenegro

Safetak : Pthread biblioteke se koriste za paralelizaciju aplikacija pisanih u C -u. Kreirane niti se koriste za simultano izraiunavanje kompleksnib statistidcili proracuna. Parametri Weibull distribucije se procjenjuju koriscenjem metoda maksimalne vjerovatnoce, koji zahtijeva veliki broj matematic"kih operacija. U radu ce biti prezentovano kako dodavanje Pthread koda u aplikaciju pomaie ubrzanju izvrsavanja paralelizacijom niti.

Abstract: The Pthread library is used in C applications to make it in a parallel version. The threads created are used to compute simultaneously the calculations in complex statistical formulas. This is a solution in cases when time of execution of such application is too long. Using parallel threads can lead us to a faster result. One of these is calculating the parameters of Wei bull distribution when the sample data is given. These parameters are estimated by using the maximum likelihood method, which requires many mathematical computations. This article shows how the Pthread code added in the application helps to perjormfaster by doing it in parallel threads.

1. INTRODUCTION

Using the technique of parallelizing the code by two different processors that use the same merrory space, it makes possible that complex tasks might be computed in a shorter time as they were computed by a single processor architecture. Pthreads are a set of types and procedure calls in C. They are implemented in a pthread.h file, and a thread library. [1] The POSIX thread library provides an interface to generate and interact with separate threads of execution within a program. This standard is defined by the IEEE and is available across nearly all variants of the UNIX operating systems. The processor creates threads that are used to compute a part of calculations of the whole application. They are running concurrently, so the total time to compute the result is smaller and is depending on the nunoer of threaded created by the processor. The primary rrotivation for using threads is to realize potential program performance gains. Comparing to processes, creating a thread requires fewer operations, and to manage a thread requires fewer system resources. Another advantage is related to the software portability. Applications that use threads can be developed on serial machines and run on parallel machines without changing anything. This portability is very significant advantage of threaded APIs. In this article is presented a project that uses Pthreads to calculate the weibull distribution parameters when data samples are given. There are calculated the speed-up in cases when the number of threads increases.

2 WEIBULL DJSTRffiUTION

The weibull distribution is an important distribution especially in maintainability analysis and reliability. Furtherrrore it is widely used in biomedical engineering and

to model variations in wind speed. This distribution exists in two main forrrs: the two-parameter and three-parameter Weibull distribution. The two parameter Weibull Distribution has the following density and distribution functions:

[(X) = (~) (~t-l e -(~t (1)

F(x)=l-e-(~t (2)

where a is the scale parameter and b is the shape parameter. The three parameter Weibull distnbution has the following formula of density and distribution:

[(X) = (£)t-c) b-le _(x:ct a a

(3)

F(x) = 1 _ e _(:C)b (4)

where a is the scale parameter, b is the shape parameter and c is the location parameter. The application makes an estimation of the parameters of the Weibull distribution function. Two main methods that can be used are the maximum likelihood and the least squares. In the application it is used the maximum likelihood method, as it is usually considered to be rmre robust and produces more accurate res ults. The method of maximum likelihood [2] is a comrronly used procedure because it has very desirable properties. Let Xl , X2 , ••• , Xn be a random sample of size n drawn from a probability density function fx (x;9) where 9 is an unknown parameter. The likelihood function of this random sample is the joint density of the n random variables and is a function of the unknown parameter. Thus

L=TII!"l [xi(xi,9) (5)

is the likelihood function. The maximum likelihood estimator (MLE) of e , say 'o, is the value of e that maximizes Lor, equivalently, the logarithm of L . Often, but not always, the MLEof 8 is a solution of

dlogL =0 de (6)

where solutions that are not functions of the sample values x}, X2, ••• , Xn are not admissible, nor are solutions which are not in the parameter space. The problem with the MLE lies in obtaining initial estimates for x and 9. MLE does not supply the values for x and e, those values are supplied by a parameter estimation technique. The rmst COIDlllJn technique for parameter estimation of Weibull distributed reliability data is the Newton-Raphson algorithm (NRA). The Newton-Raphson algorithm uses a Taylor Series expansion about the predefined function to estimate the parameters' values. However, the main problem with the NRA algorithm is the selection process for the initial parameter values. The selection of the initial paraneter values is critical, because of the inability of the NRA to avoid convergence to local optima. f(x; 8) is the function which is trying to be optimized. In this research the function f(x; e), is the Two-Parameter Weihull Distrihution probability density function.

3.ALGORITHM FOR WEIBULL DISTRmUTION

The program takes a censored set of data, which might be ordered or unordered, taken from a sample of N data. The location parameter A is given, and the algorithm makes an estimation of the scale parameter B and shape parameter C of the equation (4) of the Weibull distribution. The method used for this estimation is the maximu m likelihood estimation. To calculate the root of(6) where the L is the likelihood function and is computed via a Newton-Raphson iteration. The Newton-Raphson method in one variable is implemented as follows: Given a function f defined over the reals x and its derivative f " we begin with a first guess Xo for a root of the function f [3]. The function is reasonably well-behaved a better approximation Xl is

(f.ro) ,rl = ·1'0 - ;-'(.1'0) .

(7)

In our case, the value of Xo is given, and if it is zero is it taken a guess value of x. Also forthe density function is calculated the node, the mean and the variance for the given set of data. This procedure is repeated for R times for different set of data, taken from this N data sample. The algorithm is written in C, including the Pthread library. The system where it is executed is a grid cluster, a Linux based system and portable enough to run with consistent result on any POSIX system The data sample and the location parameter is read from a file by each thread that is created. These parallel threads take the set of data M, from the sample N, and it is repeated Rlnr_threads times. The pseudocode for the serial version is: For each Ri times

Calculate the Weibull parameters for Mi data sample End

The pseudocode for the parallel version is : For each thread For each Rilnr _threads

Calculate the Weihull parameters for Mi data sample End

End

4.RI<SULTS OF PERFORMANCE TFSTS

The data sample used for testing contains 2000000 numeric data. The Weibull parameters are calculated for 100 such samples, and for each of them is calculated also the mode, the mean and the variance of the distribution. The project is tested for different number of threads, beggining with the serial version with only one thread running a doing the calculations, to eight threads in parallel doing the calculations. In the table land table 2 are given the results for the time in seconds and for the speed-up for different number of threads used.

Table 1: Performance in lime and speed up of the applicalion with a sample of2 000000 data.


1 2436 1

2 816 2.9

3 537 4.5

4 408 5.9

5 325 7.4

6 260 9.3

7 225 10.8

8 187 13.0

Table 2: Performance in lime and speed up of the application with a sample of 3 000 000 data.


1 2466 1

2 1246 1.9

3 784 3.1

4 605 4.0

5 531 4.6

6 458 5.3

7 360 6.8

8 321 7.6

The figure 3 and figure 4 show a graphical presentations for the tin-e spent for different number of threads. Time in reduced immediately as the second thread is added, increasing in this way the performance of all the parallel executions ofthe application.

'. .


3000

2500

';2000 .. ~ 1500 .! • !looo

500

0 0 2 4 6 8 10

Number of threads


4 6 8 Numberofthreads -~~ ..

The figure 5 shows the speed-up graphic for these performance tests. The tests shows that in both cases the speed up increases in a linear form.

Figure 5:Graphical presentation of the performance of speed up 14

12 +:-:==-..:......-+-......;::::;.=F.::..;=::;=;=f-=-Er---f=~::1

g 8 ~~---+~~~~~~~~~~ __ --~~ 16~~~~~~~~~~~~ ..

o 4 6 8 10

Number ofthreads

__ Sample with 2 000 000 data

~Samplewith 3 000000 d.",

5.CONCLUS IONS

This paper introduced the benefits of using the Pthread library as a function of highlighting the advantaged of the parallel versions in mathematical and statistical calculations. The Pthread has the ability to create concurrent excecutions instances on a single processor. It also allows to share global memory and maximizes the machine's resources.

10

Also it is worth noting the difficulty that is required to tum the serial code into a parallel code using the threads. The programmer has to make this explicitly in the code, by using the Pthreads library functions of creating and managing them It is required rmre work to the programmer to know how the threads work and how they will share the memory between them The programmer should know how to manage the access of a shared variable in a simultaneously way. Also the programmer should be careful with the grobal variables . Threads should not make changes to them, but only by using the mute x semaphore. The cost of creating threads, instead of using only one thread in the serial version, is the corrplexity of the code, and the effort of the programmer to make it work without deadlocks. However, co npared to the cost of creating and managing a process, a thread can be created with much less operating system overhead. Managing threads requires fewer system resources than managing processes. As a final result, the miltithreading way of paralleli:ze is very worthy in case when different calculations can be computed independently by the same processor. Another way of parallelize is by using the message passing interface (MPI) which is used to exchange messages between different processors, in a multiprocessor system As future work might be testing the performance of a parallel hybrid version of using the MPI and Pthreads both to make calculations of the parameters of a Weibull distribution for some data samples.

REFERENCES

[I] B. Lewis,D.J. Berg, (2008), Pthreads Primer.

[2] Harter, H. L. and Moore, A. H. (1%5). Point and Interval Estimation, Based on m-order Statistics, for the scale parameter of a Wiebull Population with Known Shape parameter. [3] http://en.wikipedia.org/wikilNewton%27s_method

Curriculum Vitae

Personal information First name I Surname Sid ita Duli

Address Street: "Vellezerit Frasheri " Nr 42. Shkoder, Albania Telephone +35522241550 Mobile: +355692644356

[email protected] [email protected]

Web site : \wNj.siditaduILcom

Nationality Albanian

Date of birth 22.08.1983

Gender Female

Work experience Dates October 2006- now

Occupation or position held Lecturer, Department of Mathematics and Informatics, University "Luigj Gurakuqi" Shkoder,Albania Main activities and responsibilities Subjects I have taught during these years:

In Computer Sciences classes: Java, Operating Systems, Web Applications In Business Administration classes: Basics of computer sciences In Master classes: Scratch Programming

Name and address of employer University "Luigj Gurakuqi" Shkoder, Address: Sheshi "2Prilli" Shkoder, Albania Type of business or sector University

Dates March 2015 - April 2016 Occupation or position held Vice-Dean of Faculty of Natural Sciences

Main activities and responsibilities Administrative tasks in accreditation process of the faculty. Name and address of employer University "Luigj Gurakuqi" Shkoder, Address: Sheshi "2Prilli" Shkoder, Albania

Type of business or sector University

Dates December 2010 - December 2014 Occupation or position held iOS developer, member of team Internetpeople in Shkoder, Albania

Main activities and responsibilities iOS application development, Java application development, web development tasks Name and address of employer Internetpeople SHPK

Type of business or sector Software development

Education Dates

Title of qualification Principal subjects

Name and type of organisation providing education and training

July 2010- now ~ PhD Candidate Thesis in "Parallel implementation of parameter estimation o~ Weibull distributi~n parameters" - Electro-technical Faculty, University of Montenegro

Level in national or international PhD studies classification

Dates October 2006- June 2008 Title of qualification Master in Informatics Principal subjects Database

Operating Systems Parallel Computing Theory of coding and cryptography Master thesis: Web services which enable data exchange between different applications

Name and type of organisation Faculty of Computer Sciences, University of Tirana, Albania providing education and training Level in national or international Master degree, 2 years

classification

Dates October 2002- July 2006 Title of qualification Specialist in Informatics Principal subjects Programming languages ( Java, C++, Visual Basic, PHP)

Database Operating Systems Web Application Mathematical Analysis

Name and type of organisation Faculty of Computer Sciences, University "Luigj Gurakuqi" Shkoder, Albania providing education and training Level in national or international University degree, 4 years

classification

Trainings

Personal skills and competences Mother tongue

Other languages

Self-assessment European level

English Italian

German

October 2010 Attend the 9-nth SimLab course in Belgrade, Serbia September 2005 Attend the qualification courses of the project" Tempus" for training new teachers.

Albanian

~

English, Italian, German Understanding Speaking Writing

Listening Reading Spoken interaction Spoken production

C1 Proficient user C1 Proficient user C1 Proficient user C1 Proficient user C1 Proficient user - - - C2 Proficient user C2 Proficient user C2 Proficient user C2 Proficient user C1 Proficient user

91 Independent user Bl Independent user Bl Independent user B1 Independent user B1 Independent user

Certificates

The English certificate from University of Tirana, level cz, obtained in April 2006 Two German certificates from University "Alpen-Adria" Klagenfurt, Austria, during two summer courses Obtained in September 2005, level A2, and September 2006 level B1

------1

Social skills and competences I am a social person, good ability in communication, good in working in a team.

Organisational skills and Good in leadership and in organising a team, especially during the workshops with students. competences

Technical skills and competences Very good in Java programming, C++.

Also, 4 years experience in mobile application development.

Additional information Publications:

Papers in scientific journals • S.Duli, B.Krstajic "Parallel implementation of the Weibull distribution parameters estimator", in Journal of Environmental Protection and Ecology (JEPE), ISSN 1311-506, March 2014 • .5..Q.u.li, B.Krstajic "Parallel processing of wind speed data during years 2012 _ 2013 in Shkodra region" in the Scientific Bulletin of Faculty of Natural Sciences, University of Shkodra, volume 64. March 2014 • S.Duli, B.Krstajic "Parallel computing of Weibull distribution parameters" in the Scientific Bulletin of Faculty of Natural Sciences, University of Shkodra, volume 62. ISSN : 2221-6847, http://80.78.68.65:1000/buletin/Ol_BuI_62.pdf, March 2012 • S.Duli, I. Ninka, "Web Service qe realizon sbkembim te dhenash nen protokollin SOAP" ("Web Services which enables data exchange between different applications") in the Scientific Bulletin of Faculty of Natural Sciences, University of Shkodra, volume 59. June 2009

Papers in scientific conferences • S.Duli, B.Krstajic, "Hybrid MPIiPthread parallelization of the Weibull distribution parameters estimator", XIX Scientific-Professional Information Technology Conference, Zabljak, February 2014 • S.Duli, B.Krstajic "Methods for estimating the parameters of the Weibull distribution" in 2nd

International Conference "Research and education in natural sciences", ISBN 978-9928-4135-5-0, November 2013 • S.Duli, B.Krstajic "MPI in a Weibull distribution parameters estimation" International Conference "Towards future sustainable development", ISBN 978-9928-4135-0-5, November 2012 • S,Duli, B.Krstajic, "Pthread u estimaciji parametara weibull distribucije", XVII Scientific Professional Information Technology Conference, Zabljak, February 2012 • .s..Qilli, B.Krstajic, "Parallel database processing approaching", International Conference "Challenges of European Economic Integration of Western Balcan" ISBN 978-9928-4011-2-0, Shkoder, Albania, December 2010 • SJMi, B.Krstajic, "PRIMJER VALIDACIJE XML-a POMOCU DTD-a", XV Scientific-Professional Conference Information Technology, Zabljak, February 2010 • SJ2l.!!i, "Web services authentification and data security", International Conference "Economies in transition - during and after" , ISBN 978-99956-02-81-9, Shkoder, Albania, December 2009 • S.Duli, K.Nikaj, E.Gavo<;i, "Rich internet applications: an implementation in Java Web Start in e learning", 4th International Conference "Harmonisation of Research and Education with Sustainable Development", November 2017 • K.Nikaj, S.Duli, E.Gavo<;i, "Using computer simulations in teaching and learning nuclear physics at high school", 4th International Conference "Harmonisation of Research and Education with Sustainable Development", November 2017 • S.Duli, K.Nikaj, E.Gavoc;i, "Rich internet applications: a comparison between java applets and other similar platforms", 3rd International Conference "Research and Education in Natural Sciences", Nentor 2015 • E.Gavoc;i, K.Nikaj, S.Duli, "The effectiveness of integration of interactive physics simulations in secondary school: topics from electrical physics", 3rd International Conference "Research and Education in Natural Sciences", Nentor 2015

Projects: 2007 - Member of project for building and maintaining the official web site for the University of Shkodra 2011 .; Coordinator of the e-Iearning project, Faculty of Natural Sciences, University of Shkodra

ponofoglio as software developer -Parallel implementations in C and Fortran (using MPI and Pthreads) -Websites using Joomla CMS and Php (as part of the team InternetPeople) -iOS apps : Barcode, Fuel Mix Calculator, (as part of the team InternetPeople) -Google maps projects (as part of the team InternetPeople)

Bibliography

SiditaDuli

PhD candidate

Papers in scientific journals

• S.Duli, B.Krstajic "Parallel implementation of the Weibull distribution parameters estimator", in Journal of Environmental Protection and Ecology (JEPE), ISSN 1311-506, March 2014

• S.Duli, B.Krstajic "Parallel processing of wind speed data during years 2012 - 2013 in Shkodra region" in the Scientific Bulletin of Faculty of Natural Sciences, University of Shkodra, volume 64. March 2014

• S.Duli, B.Krstajic "Parallel computing of Weibull distribution parameters" in the Scientific Bulletin of Faculty of Natural Sciences, University of Shkodra, volume 62. ISSN: 2221-6847, http://80.78.68.65:1000Ibuletin/OLBuL62.pdf, March 2012

• S.Duli, I. Ninka, "Web Service qe realizon shkembim te dhenash nen protokollin SOAP" "Web Services which enables data exchange between different applications" in the Scientific Bulletin of Faculty of Natural Sciences, University of Shkodra, volume 59. June 2009

Papers in scientific conferences • S.Duli, B.Krstajic, "Hybrid MPI/Pthread parallelization of the Weibull distribution

parameters estimator", XIX Scientific-Professional Information Technology Conference, Zabljak, February 2014

• S.Duli, B.Krstajic "Methods for estimating the parameters of the Weibull distribution" in 2nd International Conference "Research and education in natural sciences", ISBN 978- 9928-4135-5-0, November 2013


1


• S.Duli, B.Krstajic, "Pthread u estimaciji parametara weibull distribucije", XVII Scientific-Professional Information Technology Conference, Zabljak, February 2012

S.Duli, B.Krstajic, "Parallel database processing approaching", International Conference "Challenges of European Economic Integration of Western Balcan" ISBN 978-9928- 4011-2-0, Shkoder, Albania, December 2010

S.Duli, B.Krstajic, "PRIMJER VALIDACIJE XML-a POMOC:U DTD-a", XV Scientific-Professional Conference Information Technology, Zabljak, February 2010

•

•

• S.Duli, "Web services authentification and data security", International Conference "Economies in transition - during and after" , ISBN 978-99956-02-81-9, Shkoder, Albania, December 2009

• S.Duli, KNikaj, EiGavcci, "Rich internet applications: an implementation in Java Web Start in e-learning", 4th International Conference "Harmonisation of Research and Education with Sustainable Development", November 2017

• KNikaj, S.Duli, E.Gavo<;;i, "Using computer simulations in teaching and learning nuclear physics at high school", 4th International Conference "Harmonisation of Research and Education with Sustainable Development", November 2017

• S.Duli, KNikaj, E.Gavo<;;i, "Rich internet applications: a comparison between java applets and other similar platforms", 3rd International Conference "Research and Education in Natural Sciences", Nentor 2015

• EiGavoci, K.Nikaj, S.Duli, "The effectiveness of integration of interactive physics simulations in secondary school: topics from electrical physics", 3rd International Conference "Research and Education in Natural Sciences", Nentor 2015

2

Biografija - Milutin Radcnllc

Prof. dr Milutin Radonjlc je roden 9.07.1966. godine u Beogradu. Osnovnu i srednju skolu zavrsio je u Titogradu, stekavsl zvanje pornocnog lstraflvaca u matematici. Kao ucenik osnovne skole ostvario je zaoazene rezultate na takrnicenju iz fizike u okviru pokreta "Nauka mladima". Nosilac je dip lome "Luca",

Studije na Elektrotehnlckorn fakultetu zavrslo je 1991. godine, sa prosjecnorn ocjenom 9,56. Kao student cetvrte godine nagraden je studentskom nagradom "19. decembar". Za vrijeme studija bio je korisnik stipendije Vlade Republike Crne Gore za talentovane studente. Od februara 1993. godine do danas radi na Elektrotehnlckorn fakultetu u Podgorici u zvanju asistenta pripravnika, asistenta, docenta i vanrednog profesora.

Postdiplomske studije zavrslo je 1997. godine na ETF-u u Podgorici, na odsjeku za Racunare, sa prosjecnorn ocjenom 10. Doktorsku tezu pod naslovom "Prilog analizi performansi CQ komutatora paketa sa stanovlsta vellclne i algoritama upravljanja redovima cekanja" odbranio je 19. maja 2011. godine na Elektrotehnlckom fakultetu u Podgorici.

U septembru 2012. godine izabran je u zvanje docenta, a u oktobru 2017. godine u zvanje vanrednog profesora na Elektrotehnlckorn fakultetu Univerziteta Crne Gore.

Oblasti njegovog naucnog interesovanja su: mikroprocesorski sistemi, racunarske rnreze, projektovanje digital nih sistema.

Kao autor iii koautor objavio je sedam radova u referentnim medunarodnim casoptstrna sa SClliste, vise radova u regionalnim i dornaclm casoptslma, vise od sezdeset radova na medunarodnim i regionalnim konferencijama i dva rada po pozivu na naucnim skupovima. Autor je zbirke zadataka u izdanju Univerziteta Crne Gore i koautor jednog udzbenlka.

Clan je IEEE i ACM. Clan je i programskog odbora konferencije "Informacione tehnologije", uredivackog odbora casopisa "ETF Journal of Electrical Engineering" i recenzent u vise referentnih medunarodnih casopisa. Clan je tehnickog komiteta za informacione tehnologije u Institutu za stand ardizaciju Crne Gore.

Ucestvovao je kao clan projektnog tima na po jednom medunarodnom COST i IPA projektu, na dva medunarodna projekta finansirana od strane EU kroz FP7 program, na dva bilateralna projekta, na dva nacionalna projekta i na prvom centru izvrsnosti u Crnoj Gori (BIO-ICT). Ima i visego disnje uspjesno iskustvo u saradnji sa privrednim subjektima na mnogobrojnim projektima.

Za vrijeme rada na medunarodnim projektima imao je krace studijske boravke na University of Ghent (Belgija) i u kompaniji Erikson Nikola Tesla (Hrvatska).

Od septembra 2013 godine do dan as obavlja funkciju prodekana za razvoj i lstrazivan]e na Elektrotehnlckom fakultetu Univerziteta Crne Gore.

Najvaznije reference - Milutin Radonjic

Casopis! sa SCI/SCIE Iiste:

• Gardasevic G., Veletic M., Maletic N., Vasiljevic D., Radusinovic I., Tomovic S., Radonjic M., "The

loT Architectural Framework, Design Issues and Application Domains", Wireless Personal Communications, Vol 92, No 1, January 2017, pp. 127-148, ISSN: 0929-6212, (print version), ISSN: 1572-834X (Online). 001: 10.1007/s11277-016-3842-3

bJt:p:ilmil.darivate.com/cgi pioIirDI:>..:tL!lresylts.CRt?PC=MASTJR$<Full=Wireless%20Per2on..?I%f.Q._C;_Qn:ml~micatiq.[L:>"

• Radonjic M., Ljumovic N., Misovic D., Maljevic I., Yoshigoe K., Radusinovic I., "CQ Ethernet Switch Implementation on the NetFPGA Platform", Wireless Personal Communications, Vol 92, No 1, January 2017, pp. 5-19, ISSN: 0929-6212, (print version), ISSN: 1572-834X (Online). 001: 10.1007/s11277-016-3835-2

i!l:tp:/.i!.D.i1&Jariva!e.com_igsj: ginlirnJ5..tLi!!esults:cgi'!)t?C::::IY1ASTER&Full::::Wireless%20Person~OCo[l1.J.rl.\:lDJ.~il..E9.n.~

• Gardasevic G., Divanovic 5., Radonjic M., Radusinovic I., "A QoS-aware Dual Crosspoint Queued switch with Largest Weighted Occupancy First scheduling algorithm", IEICE Transaction on Communications, VoI.E98-B, No.Ol, January 2015, pp. 201-208, ISSN: 0916-8516 (print version), ISSN: 1745-1345 (Online). 001: http://doi.org/10.1587/transcom.E98.B.201

IlEPJiJ'ljl.c1arivate.com/cgi-bin/irnist!jiresults.cgi'?PC=MASTER&ISSN:::::0916-851§

• Radusinovic I., Radonjic M., Simurina A., Maljevic I., Veljovic Z., "A new analytical model for the CQ switch throughput calculation under the bursty traffic", International Journal of Electronics and Communications (AEU), Vol. 66, No 12, December 2012, pp.1038-1041, ISSN: 1434-8411. 001: http://dx.doi.org/10.1016/j .aeue.2012.05.009

!}:\\p:jj_p,jl,darivate.com!cgi-bin!jrnlst!jlresults.cgi·?PC=MASTER&ISSN::::1434-84J.1

• Radonjic M., Radusinovic I., Simurina A., Banovic D., "A New Analytical Model for the CQ Switch Performance Analysis under the Bursty Traffic", IEICE Transaction on Communications, Vo1.E95- B, No.2, February 2012, pp.595-598, ISSN: 0916-8516 (print version), ISSN: 1745-1345 (Online). 001: http://dx.doi.org/10.1587 /transc om.E95.B.595

f1:up.Jln.lil.darivate.com!cgi-binLjrnlst!llresults.cgi?PC=MASTER&IS?N=0916 .. 851§

• Radonjic M., Radusinovic I., "Impact of scheduling algorithms on performance of crosspoint queued switch", Annals of Telecommunications, Vol 66, No 5-6, May/June 2011, pp.363-376, ISSN: 0003-4347 (print version), ISSN: 1958-9395 (electronic version). 001: 10.1007/s12243-010- 0214-y

http://mjl.c1arivate.com/cgi-bin/jrnlstjjlresults.cgi?PC=M ASTER&I SSN =0003-4347

• Radonjic M., Radusinovic I., Cvorovic J., Yoshigoe K., "Iterative throughput calculation for crosspoint queued switch", IEICE Transactions on Communications, Vol E93-B, No 12, December 2010, pp. 3635-3638, ISSN: 0916-8516 (print version), ISSN : 1745-1345 (Online). 001: http://dx.doi.org/10.1587 /transcom.E93.B.3635

rl!tp:fLITIiL~.I?_[j.'@te. comJ.5,:_gj::.RlDIl!:nl?lLilr.~~.l)!.ts. c_gj]e£:= M 65~r~8_& ISS .N..=Q.~J§=~.~.l§

Radovi u mec:iunarodnim casoptslma koji nisu na SCI/SCIE listi:

• Radonjic M., Radusinovic I., Maljevic I., Banovic D., "CQ Switch Analysis under Traffic Overload", Telfor Journal, Vol.3, No.1, 2011., pp. 19 - 22, ISSN: 1821-3251.

• Radonjic M., Radusinovic I., Maljevic I., "Packet Delay Variation Analysis of the CQ Switch", Telfor Journal, Vol.4, No.1, 2012., pp. 8-13, ISSN: 1821-3251.

Radovi u SCOPUS-u:

• Radonjic M., Radusinovic I., "Buffer length impact to crosspoint queued crossbar switch performance", Proc. of 15th IEEE Mediterranean Electrotechnical Conference (Melecon 2010), pp. 119-124, Valletta, Malta, April 2010. 001: 10.1109/MELCON.2010.5476326

• Radonjic M., Radusinovic I., "Buffer Length Impact to 32x32 Crosspoint Queued Crossbar Switch Performance", Proc. of 15th IEEE symposium on Computers and Communications (ISCC 2010), pp. 954-959, Riccione, Italy, June 2010. 001: 10.1109/ISCC.2010.5546762

• Radonjic M., Raduslnovlc I., "Average Latency and Loss Probability Analysis of Crosspoint Queued Crossbar Switches", Proc. of 52nd International Symposium ELMAR-2010, pp. 203-206, Zadar, Croatia, September 2010.

• Radonjic M., Radusinovic I., Veljovic Z., Maljevic I., "Performance evaluation of Crosspoint Queued Switch Under the Heavy Traffic", Proc. of 16th IEEE symposium on Computers and Communications (ISCC 2011), pp. 943-949, Corfu, Greece, June 2011. 001: 10.1109/ISCC.2011.5983963

• Misovic D., Ljumovic N., Radonjic M., Radusinovic I., "Implementation of the Crosspoint-Queued Switch's Output Controller on the NetFPGA Platform", Proc. of 53rd International Symposium ELMAR-2011, pp. 235-238, Zadar, Croatia, September 2011.

• ljumovic N., Mlsovlc D., Raduslnovlc I., Radonjic M., Banovic D., "Buffer size impact on the CQ Ethernet switch performance", Proc. of 19th Telecommunication Forum TELFOR 2011, pp. 186- 189, Belgrade, Serbia, November 2011. 001: 10.1109/TELFOR.2011.6143522

• Radonjlc M., Raduslnovlc I., Banovlc D., Maljevlc I., "Packet delay variation analysis of the CQ switch under uniform traffic", Proc. of 19th Telecommunication Forum TELFOR 2011, pp. 190- 193, Belgrade, Serbia, November 2011. 001: 10.1109/TELFOR.2011.6143523

• Radonjic M., Radusinovic I., Maljevic I., "Packet Delay Varia nce Analysis of the CQ Switch Under the Unbalanced Traffic", Proc. of 16th IEEE Mediterranean Electrotechnical Conference (Melecon 2012), pp. 1005-1008, Yasmine Hammamet, Tunisia, March 2012. 001: 10.1109/M ELCON .2012.6196597

• Divanovic S., Kovacevic V., Radonjic M., Yoshigoe K., Radusinovic I., "Crosspoint Queued Switch Performance Analysis Under Multicast Traffic", Proc. of 20th Telecommunication Forum TELFOR 2012, pp. 226-229, Belgrade, Serbia, November 2012.001: 10.1109/TELFOR.2012.6419187

• Radonjic M., Maljevic I., Lekic N., Radusinovic I., "Performance Analysis of Variable Packet Size Crosspoint-Queued Switch", Proc. of IEEE Eurocon 2013, pp. 673-678, Zagreb, Croatia, July 2013. 001: 10.ll09/EU ROCON.2013.6625053

• Maletic N., Divanovic S., Radonjic M., Radusinovic I., Gardasevic G., "Performance Evaluation of Dual Crosspoint Queued Crossbar Packet Switch", Proc. of TELSIKS 2013, pp. 145-148, Nis, Serbia, October 2013. DOl: 10.1l09/TELSKS.2013.6704910

• Divanovic S., Radonjic M., Gardasevic G., Radusinovic I., "Dynamic Weighted Round Robin in Crosspoint Queued Switch", Proc. of 21th Telecommunication Forum TELFOR 2013, pp. 109-112, Belgrade, Serbia, November 2013. DOl: 10.1109/TELFOR.2013.6716184

• Rad usinovic I, Divanovic 5., Radonjic M., "Analysis of WRR Scheduling Algorithm Frame Size Impact on CQ switch Performance", Proc. of 17th IEEE Mediterranean Electrotechnical Conference (Melecon 2014), pp. 403-407, Beirut, Lebanon, April 2014. DOl: 10.1109/M ELCON .2014.6820568

• Zaric N., Radonjic M., Kyriazakos S., Pejanovic-Djurisic M., "Automated Algorithm for Classification of Water-flow Signals to Support Ambient Assisted Living Applications", Proc. of 22nd Telecommunication Forum TELFOR 2014, pp. 31-34, Belgrade, Serbia, November 2014.

• Zaric N., Radonjic M., Pejanovic-Djurisic M., Radusinovic I., "An Example of Monitoring System with Reasoning Module for Ambient Assisted Living Application", Proc. of IEEE Eurocon 2015, pp. 30-35, Salamanka, Spain, September 2015.

• Tomovic S., Radonjic M., Radusinovic I., "Bandwidth-Delay Constrained Routing Algorithms for Backbone SDN Networks", Proc. ofTELSIKS 2015, pp. 227-230, Nis, Serbia, October 2015.

• Savic T., Radonjic M., "One Approach to Weather Station Design Based on Raspberry Pi Platform", Proc. of 23rd Telecommunication Forum TELFOR 2015, pp. 623-626, Belgrade, Serbia, November 2015.

• Savic T., Radonjic M., "Proposal of Solution for Automated Irrigation System", Proc. of 24th Telecommunication Forum TELFOR 2016, pp. 647-650, Belgrade, Serbia, November 2016.

Univerzitet erne Gore adresn / address __ Cetinjska hr. 2 81000 Podgonca, Crna Cora tclefim ! phone _00382 20 414 255 fax_ 00382 20 414 230

Broj I Ref 03- 2Gb.5 Datum I Date J~. 10. 20.11_

[email protected] wcb_www.ucg.ac.me

University of Montenegro

Na osnovu clana 72 stav 2 Zakona 0 visokom obrazovanju ("Sluzbeni list erne Gore" br. 44/14, 47/15,40/16,42/17) i clana 32 stav 1 tacka 9 Stat uta Univerziteta erne Gore, Senat Univerziteta erne Gore na sjednici odrzano] 16.oktobra 2017.godine, donie je

ODLUKU o IZBORU U ZVANJE

Dr Milutin Radonjlc bira se u akademsko zvanje vanredni profesor za oblast Digitalni sistemi i informatika na Elektrotehnickom fakultetu, na period od pet godina.

Senat Univerziteta erne Gore Predsjedavaju ' .--,-

Danilo Nikolic,v.f.rektora

Prof. dr Bozo Krstajic, redovni profesor

• Kratka biografija

Roden je 7. aprila 1968. god. u Zabljaku, gdje je zavrslo osnovnu skolu i prva dva razreda srednjeg usmjerenog obrazovanja. Srednju skolu je zavrsio u gimnaziji "Slobodan Skerovic" u Podgorici. Na Elektrotehnickom fakultetu u Podgorici je diplomirao marta 1992. godine sa prosjecnorn ocjenom 9,87, a diplomski rad "YAMABICO - upravljanje mobilnim robotorn" je odbranio sa ocjenom 10. Dobitnik je studentske nagrade "19. decembar" i Plakete Univerziteta kao najbolji student Univerziteta 1991. god. Postdiplomske studije je upisao na istom fakultetu 1992. godine, na Odsjeku robotike i vjestacke inteligencije. Ispite na postdiplomskim studijama je polozio sa prosjecnom ocjenom 10, a magistarski rad pod nazivom "Modifikovani adaptivni LMS algoritmi" je odbranio 1996. godine. Doktorsku disertaciju, pod nazivom "Novi pristup LMS adaptivnom algoritmu sa promjenljivim korakorn", odbranio je 20. 12. 2002. godine na Etektrotehnickom fakultetu u Podgorici.

U zvanje docenta je izabran 09.07.2003. godine, u zvanje vanrednog profesora 02.10.2008. godine, a zvanje redovnog profesora 19.12.2013. godine na Univerzitetu Crne Gore. Bio je visiting profesor od 2004. do 2007. godine na univerzitetu "Luigi Gurakuqi" u Skadru, Albaniji. Bio je direktor Centra informacionog sistema UCG od 2003. do 2015. godine.

Autor je iii koautor dvije monografije, vise udzbenika za osnovnu skolu iz oblasti informatike i vise autorizovanih skripti za potrebe nastave na predmetima na kojima je anqazovan. Do sada je objavio preko 100 naucnih i strucnih radovan u casopisima i na konferencijama. Mentor je na 2 doktorskom radu i 5 magistarskih radova, a pod njegovim mentorstvom su uspjesno zavrsena: 2 doktorska, 10 magistarskih i preko 150 diplomskih i speciia'istickih radova. Recenzirao je vise naucnih radova u istaknutim svjetskim casoplsima iz oblasti adaptivnih algoritama i racunarsklh sistema.

Koordinirao je i ucsstvovao u vise znacajnih evropskih projekata kao predstavnik Univerziteta Crne Gore, a koje finansira Evropska unija u okviru FP6, FP7, TEMPUS, IPA i H2020 programa (SEEREN2, SEE-GRID2, SEE-GRID-SCI, SEERA-EI, GEANT3, NQF&QHE, GEANT3+, HPSEE, EGI-Inspire, DL@WEB, RINGINDEA, FORSEE, CONGRAD, GN4 i VI-SEEM). Anqazovan je od strane vise kompanija i institucija u Crnoj Gori ivan nje kao strucni ICT konsultant iii projektant, te je projektovao i realizovao vise znacajnih strucnih projekata. Od strane sudova u Crnoj Gori je anqazovan kao sudski vjestak za oblast ICT-a.

Predsjednik je organizacionog i programskog odbora dornaceq naucno-strucnoq skupa »INFORMACIONE TEHNOLOGIJE« koja se vee 23 godine organizuje i editor je zadnjih 9 zbornika ove konferencije. Takodje je clan programskih odbora dvije medunarodne konferencije: "Balkan Conference in Informatics" i "RoEduNet Conference: Networking in Education and Research" kao i clan Predsjednistva ETRAN-a. Clan je medunarodne asocijacije elektro inzeniera - IEEE, lnzenjerske komore Crne Gore, Internet zajednice ISOC i Ubuntu zajednice Crne Gore. Menadzer je lokalne CISCO akademije.Pokretac je MREN-a (Montenegrin Research and Education Network) i clan njegovog upravnog odbora. Osnivac je prvog IXP-a u Crnoj Gori (MIXP). Govori engleski jezik, a stuzi se i ruskim jezikom.

• Najvaznije i najsvjezije reference iz oblasti doktorata (radovi 1,2,3,4, 5, 6 i 7 su u casopisima sa SCI liste):

1. T. Popovica, N. Latinovic, A. Pesic, Z. Zecevic, B. Krstajic i S. Djukanovic, .Architectmq an loT-enabled platform for precision agriculture and ecological

monitoring: A case study", Computers and Electronics in Agriculture, Vol. 140, August 2017, pp 255-265, ISSN 0168-1699, dOi.org/10.1016/j.compag.2017.06.008, Elsevier

2. L. Filipovi6, D. Mrdak and B. Krstajic, "Performance evaluation of parallel DNA multigene sequence analysis", Comptes rendus de l'Academle bulgare des sciences, Vol 69, No.4, 2016. pp.489-496. Print ISSN 1310-1331, Online ISSN 2367-5535.

3. Z. Zecevic , B. Krstajic and M. Radulovic, "Frequency-domain adaptive algorithm for improving the active noise control performance", lET Signal Processing, Volume 9, Issue 4, June 2015, p. 349 - 356 001: 10.1 049/iet-spr.2014.0182, Print ISSN 1751-9675, Online ISSN 1751-9683.

4. Z. Zecevic , B. Krstajic and M. Radulovic, "A new adaptive algorithm for improving the ANC system performance", AEU-INTERNATIONAL JOURNAL OF ELECTRONICS AND COMMUNICATIONS, 001: 10.1016/j.aeue.2014.11.002, (ISSN:1434-8411), publikovan online 11/2014., Elsevier

5. S. Duli, B. Krstajic, "Parallel Implementation of the Weibull Distribution Parameters Estimator", The Journal of Environmental Protection and Ecology (JEPE), ISSN 1311- 5065, Vo1.15, No 1., pp 287 - 293, 2014. SciBulCom Ltd

6. B. Krstajic, Z. Zecevic and Z. Uskokovic, » Increasing convergence speed of FxLMS algorithm in white noise environment" AEU-INTERNATIONAL JOURNAL OF ELECTRONICS AND COMMUNICATIONS, 001: 10.1016/j.aeue.2013.04.012, (ISSN:1434-8411), publikovan online 2013., Elsevier

7. T. Popovic, M. Kezunovic and B. Krstajic, »Smart grid data analytics for digital protective relay event recordings", INFORMATION SYSTEMS FRONTIERS, 001: 10.1007/s10796-013-9434-9, (ISSN: 1387-3326, online ISSN 1572-9419),2013., Springer ..

8. L. Filipovic, D. Mrdak and B. Krstajic, "Performance Evaluation of Computational Phylogeny Software in Parallel Computing Environment", ICT Innovations 2012 Advances in Intelligent Systems and Computing, volume 207, pp 255-264, 001: 10.1007/978-3-642-37169-1_25, (ISSBN 978-3-642-37168-4, online ISSBN 978-3- 642-37169-1),2013., Springer ..

9. Lf'ilipovic, B. Krstajic, "Modified master-slave algorithm for load balancing in parallel applications", ETF Journal of ELECTRICAL ENGINEERING, vel. 20, No.1, pp. 74-83, 2014.

10. S. Duli, B.Krstajic, "Parallel computing of weibull distribution parameters", Scientific Bulletin of Faculty of Natural Sciences, University of Shkodra, volume 62, ISBN: 2221- 6847, pp. 7-14, March 2012.

J'l1HJJtPJJ1l'hJ l(PHE [OPE Y,1. UCTIUbCKa 6p. 2 n. «flux 99 81000 nOllfOPl1UA IlPHA rOPA TeJle4>oH: (020) 414·255 <l>aKc: (020) 414-230

UNIVERSITY OF MONTENEGRO

E-Illllil: rl'ktomillc.mc

VI. Cetinjska br. 2 P.O. BOX 99 81000 IJODGORICA MONTENEGRO Phone: (+382) 20 414-255 Fax: (+382) 20 414-230 E-mail: [email protected]

Bpo]: 08 - ~ loy Jl:rrYM, _ \:1 . Ii . ,20~r. I~ef:

nate, _

Na osnovu clana 75 stav 2 Zakona 0 visokom obrazovanju (Sl,list RCG, br. 60103 i Sl.list CG, br. 45/10 i 47/11) i clana 18 stav 1 tacka 3 Statuta Univerziteta Crne Gore, Senat Univerziteta erne Gore, na sjednici odrzano] 19.12.2013. godine, donio je

ODLUKU o IZBORU U ZVANJE

Dr BOZO KRST AJIC bira se u akademsko zvanje redovni profesor Univerziteta Crne Gore za predmete: Operativni sistemi, osnovne studije-ETR, Adaptivni sistemi upravllanja-epecilalisticke studije EA, Modelovanje i simulacija dinamickih sistema specllallstlcke studije EA, na Elektrotehnickom fakultetu i Automatsko upravljanje, na Masinskom fakultetu.

,-, "~ 'I l _,_ :".1 ~

1"11/1,' i'·, ,! ',:'+It-: rOPE ~->;,nFj<--;"i ~.. . ,q~?l"r~T

J 'J ~I , "', I., ',. '\..

dr Slavko Gajin Nodilova 7, 11000 Beograd

tel. +381.65.3031258

email: [email protected]

Biografski podaci Slavko Gajin je roden u Zemunu 1968. godine. Nakon zavrsetka srednje skole "Matematicka gimnazija" u Beogradu 1997. god. i sticanja diplome Vuk Karadzic, upisao je Elektrotehnicki fakultet u Beogradu (ETF), smer racunarska tehnika i informatika, nakon ceqa je odsluzio vojni rok. ETF je zavrsio u roku, 1993. godine, sa prosecnorn ocenom 9.12. Na Elektrotehntckorn fakulteta u Beogradu je odbranio magistarsku tezu pod nazivom "Analiza adaptivnosti modela zaokreta u cvrstospregnutim multiracunarskim mrezarna", a 2007, godine i doktorsku disertaciju pod nazivom "Opsti model deterministickoq rutiranja u rnultlracunarskim rnrezama".

Neposredno pred zavrsetak studija zaposlio se u Racunarskorn centru Univerziteta u Beogradu (RCUB). Krajem maja meseca 1999. god. rasporeden je na mesto zamenika direktora RCUB, a krajem 2010. godine imenovan je na mesto direktora ReUB-a,

Od 2008. godine izabran je u zvanje docenta i anqazovan je u nastavi na Elektrotehnickorn fakulteta Univerziteta u Beogradu na osnovnim, master i doktorskim akademskim studijama. Od 2009. godine anqazovan je i kao docent u nastavi na Elektrotehnickorn fakultetu u Banjoj Luci. Blo je mentor jednog svrsenoq studenta doktorskih studija.

Naucno-strucna delatnost Autor je poglavlja u knjizi medunarodnog znacaja i 33 naucnih radova ito: 7 radova u medunarodnim naucnirn casopisirna sa impakt faktorom, 13 radova prezentovanih iii objavljenih u zbornicima radova na medunarodnim naucnirn skupovima, autor poglavlja u knjizi medunarodnog znacaja, 6 predavanja po pozivu iii tutorijali na medunarodnim skupovima, 1 rad u dornacim casoplslrna, 12 radova u zbornicima radova dornacih skupova, 3 predavanja po pozivu iii tutorijali na skupovima nacionalnog znacaja.

Od 1997. godine ucestvovao je na 4 istrazivacko-tehnoloska projekta Ministarstva prosvete, nauke i tehnoloskoq razvoja, na kojima je radio najpre kao istrazivac, a kasnije i kao rukovodllac tima. Od 2002. godine ucestovao je na 10 evropskih projekata i to 8 FP projekata (FP5, FP6 i FP7), dva H2020 projekta, 1 TEMPUS i 1 INTERREG projekat, kao clan iii rukovodilac tima. Ucesnlk je sledecih tekucih medunarodnih i dornacih naucno-istrazivackih i tehnoloskih projekata:

1. H2020 GN4-phase2 - "Multi-Gigabit European Research and Education Network and Associated Services", 1.5.2015.-30.4.2016., wvv.:yY"JJeant..lls?l

2. .Prostorni, ekoloski, energetski i drustveni aspekti razvoja naselja i klimatske promene - medusobni uticaju", evidencioni broj 36035, 2011-2017. god.

Rukovodilac je sledecih projekata (navedeni su samo projekti radeni u poslednjih 5 godina, dok je kompletan spisak u prilogu):

1. Razvoj, implementacija i odrZavanje informacionog sistema za obracun i naplatu elektricne energije, RTV pretplatu i pruzanje elektrodistributivnih usluga potrosacirna, EDB, 2006-2013.

2. Matlcna evidencija osiguranih lica, RFZO, 2002-danas

3. Elektronsko fakturisanje lekova i pomagala propisanih na teret sredstava Republickoq zavoda za zdravstveno osiguranje, RFZO, 2002-danas

4. Elektronska evidencija propisanih i izdatih lekova pod specijalnim rezirnorn izdavanja, RFZO, 2002-danas

5. Zdravstveni informacioni sistem primarne zdravstvene zastite, 2005-danas 6. Razvoj, implementacija i odrzavanje Informacionog sistem za magacinsko,

materijalno i pogonsko knjigovodstvo, NBS-liN, 2009-danas

7. Razvoj, implementacija i odrzavanje informacionog sistema za pracenje proizvoda NBS-liN od posebnog interesa, NBS-liN, 2007-danas

8. Podsistem za podrsku rada pisarnice Republickoq zavoda za zdravstveno osiguranje, 2009-danas

9. Razvoj, implementacija odrzavanje Opstinskoq informacionog sistema - OpIS, 2000-danas

10. Razvoj, implementacija i odrZavanje Softverskog sistema za evidenciju maticnih knjiga, 2004-danas

11. Informacioni sistem i monitoring racunarskih rnreza - NetllS, 2002-danas

12. NetVizura NetFlow Analizer - softverski sistem za analizu rnreznoq saobracaja na bazi Cisco NetFlow tehnologije, 2013-danas

13. NetVizura EvenLog Analizer - softverski sistem za analizu syslog i snmp trap poruka, 2013-danas

14. NetVizura DNS Analizer - softverski sistem za testiranje DNS domena, 2013-danas

15. NetVizura NetFlow Monitor - softverski sistem za monitoring racunarskih rnreza, 2015-danas

Profesionalna delatnost i drustvena aktivnost Clanstvo u programskim odborima medunarodnih konferencija:

1. Member of Technical Program Committee of ICT Conference & Exhibition - INFOTECH (www.infotech.org.rs)

2. Member of Technical Program Committee of Smart E-Government International Conference and Exhibition (www.smartegov.rs)

3. Member of Technical Program Committee of Conference ICT INOVATIONS, Association for Information and Communication Technologies ICT-ACT (www.ict-act.org)

Clanstvo u medunarodnim i dornacirn strucnim telima: 1. Nacionalni predstavnik u strucnorn telu .e-Infrastructure Reflection Group" (e-IRG),

koje ima savetodavnu ulogu Evropske komisije u oblasti elektronskih infrastruktura

2. lnzenierska komora Srbije

Priznanja i nagrade 1. Plaketa Drustva za informatiku Srbije za izvanredan doprinos u razvoju informatike

u 2002. godini (projekat NetlS, RCUB).

Strana 2

Spisak radova, tehnickih resenja i projekata - doc. dr Slavko Gajin

Poglavlje u knjizi medunarodnog znacaja:

1. Slavko Gajin, Chapter 2, "Video services in Serbia's Academic Network", book: "Video Conference as a tool for Higher Education", Firenze University Press, 2012, ISBN 978-88-6655-102-7, (M14)

Radovi u medunarodnim naucnirn casoplsima sa impakt faktorom:

1. Slavko Gajin, Zoran Jovanovic, "An Accurate Performance Model for Network-on Chip and Multicomputer Interconnection Networks", Journal of Parallel and Distributed Computing, October 2012, Volume 72, Issue 10, p. 1280-1294, ISSN: 0743-7315, IF(2011)= 0.859, (M23)

2. Slavko Gajin, Zoran Jovanovic, "Explanation of Performance Degradation in Turn model", The Journal of Supercomputing, 37, 271-295, September 2006., Vol. 37, Issue 3, p. 271-295, Online ISSN 0920-8542, IF(2006)= 0.398, (M23)

3. N. Ninkovic, S. Gajin, I. Reljin, Packet Dispersion Strategy Evaluation from the Perspective of Packet Loss Pattern and VolP Quality, COMPUTER SCIENCE AND INFORMATION SYSTEMS - COMSIS, pp. 1-23,2015. (M23)

4. Y. Abuadlla, G. Kvascev, S. Gajin, Z. Jovanovic, Flow-Based Anomaly Intrusion Detection System Using Two Stages Neural Network, Computer Science and Information Systems, COMPUTER SCIENCE AND INFORMATION SYSTEMS - COMSIS, Vol. 11, No.2, pp. 601-622, Jun, 2014.(M23)

5. N. Ninkovic, Z. Bojovlc, S. Gajin, A Novel Scheme for Dynamic Triggering Of Packet Dispersion, ELEKTRONIKA IR ELEKTROTECHNIKA, Vol. 20, No.5, pp. 162-169, 2014. (M23)

6. B. Jovanovic, S. Gajin, An efficient mechanism of cryptographic synchronization within selectively encrypted H.265/HEVC video stream, Multimedia Tools and Applications (2017), pp. 1-17. https://doi.org/10.1007/s11042-017-4389-3, (M22)

7. V. Blaqojevic, D. Bojic, M. Bojovic, M. Cvetanovic, J. Dordevic, D. Durdevic, B. Furlan, S. Gajin, Z. Jovanovic, D. Milicev, V. Milutinovic, B. Nikolic, J. Protic, M. Punt, Z. Radivojevic, Z. Stanisavljevic, S. Stojanovic, I. Tartalja, M. Tornasevic, P. Vuletic, A Systematic Approach to Generation of New Ideas for PhD Research in Computing, ADVANCES IN COMPUTERS, Vol. 104, pp. 1-31, Feb, 2017 (M23)

Radovi prezentovani iii objavljeni u zbornicima radova na medunarodnim naucnirn skupovima:

1. Slavko Gajin, European Cloud Collaboration Through GEANT, Networking in Education and Research, RoEduNet IEEE International Conference, 16th edition, 21-23 September, 2017, Targu Mures, Romania (M32)

2. Valentina Tlmcenko, Slavko Gajin, "Ensemble classifiers for supervised anomaly based network intrusion detection", IEEE 13th International Conference on Intelligent Computer Communication and Processing (lCCP 2017), September 7 - 9, 2017 in Cluj-Napoca, Romania, ISBN: 978-1-5386-3367-0

3. Ibrahim Juma, Slavko Gajin, "Intrusion Detection System SON-Based, Literature review", INFOTEH-JAHORINA, Bosnia and Hercegovina, Vol. 16, March 2017, ISBN 978-99976-710-0-4

4. Marko Misic, Slavko Gajin, Koriscenje Mininet okruzenja za simulaciju softverski definisanih rnreza, 22nd Telecommunications Forum, TELFOR 2014 (25.11.2014 - 27.11.2014)

5. Slavko Gajin, Petar Bojovic: "Monitoring, analyzing and cleaning DNS configuration errors across European NRENs", TERENA Networking Conference 2013, Mastricht, Netherlands, 2013. (napomena: prihvacen rad, konferencija se odrzava 2.-6. jun 2013.), (M32)

6. Mirjana Devetakovic, Mila Pucar, Slavko Gajin, "A Knowledge Base supporting the Technological Research Project TR36035 on Climate Changes and Urban Development", ICIST 2013 - 3rd International Conference on Information Society Technology and Management, Kopaonik 2013. god., (M34)

7. M Savic, S Gajin, M Bozic, "SNMP based Grid infrastructure monitoring system", IEEE MIPRO, 2011 Proceedings of the 34th International Convention, p. 231-235, (M31)

8. Slavko Gajin, Vedrin Jeliazkov, Constantinos Kotsokalis, Yannis Mitsos: "Seamless Integration of Network Management Tools in a Multi-Domain Environment", Integrated Network Management, 2007. IM'07. 10th IFIPIIEEE International Symposium on. IEEE, 2007. p. 745-748., (M34)

9. Zoran Jovanovic, Slavko Gajin, Mara Bukvic, Pavle Vuletic, Djordje Vulovic: The Optical NREN of Serbia and Montenegro - New Solutions in Infrastructure and Monitoring, u "One step ahead", The 20th Trans European Research and Education Networking Conference, June 7-10, 2004, Rhodes, Greece, Selected Papers, ISBN 90-77559-04-3, http://www.terena.orgJpublicationslltnc2004-proceedi ngsJ, (M31)

10. Z. Jovanovic, S. Gajin, M. Bukvic, P. Vuletic, Dj. Vulovic: The optical NREN of Serbia and Montenegro, Fourth Yugoslavia-Japan Joint Workshop on Computer Simulation Science (3JW), September 2004, Tara, Yugoslavia, (M31)

11. Z. Jovanovic, S. Gajin: Network of Serbia & Montenegro, Thesalloniki, SEEREN inauguration event, January 2004., (M35)

13. Z. Jovanovic, S. Gajin: "Simulation of the Turn Model" First Yugoslavia-Japan Joint Workshop on Computer Simulation Science (3JW), 1-2 September 2000., Belgrade, Yugoslavia, (M31)

Predavanja po pozivu iii tutorijali na medunarodnim skupovima:

1. Slavko Gajin, "Monitoring and analyzing audio.video, and multimedia traffic on the network", Campus network monitoring workshop, 24-25. April 2012, Brno, Czech Republic, (M32)

2. Slavko Gajin, "ICmyNet.Flow: NetFlow based traffic investigation, analysis, and reporting", NOC Tool Workshop & 4th TF-NOC meeting, 11-12. October 2011, Brussels, Belgium, (M32)

3. Slavko Gajin, "DNS domains and servers testing", NOC Tool Workshop & 4th TF NOe meeting, 11-12. October 2011, Brussels, Belgium, (M32)

4. Slavko Gajin, "Monitoring and analyzing audio, video, and multimedia traffic on the network", NOC Tool Workshop & 4th TF-NOC meeting, 11-12. October 2011, Brussels, Belgium, (M32)

5. Slavko Gajin, "Network Monitoring System", SIRIKT 2010, 14-17. April 2010., Kranjska gora, Siovenija, (M32)

6. Slavko Gajin, "Network monitoring - NetlS", The Third CEENet Workshop on Network Management - NATO Advanced Networking Workshop "Networking the Future", 22 - 25. September 2002, Zagreb, Croatia, (M32)

Radovi u domacim casopisima;

Strana 2

1. Zoran Jovanovic, Igor Milojevic, Slavko Gajin, Milan Vitorovi6 "Sigurnost unix operativnih sistema", Infoscience 04/96, 1996. (M52)

Radovi u zbornicima radova doma6ih skupova:

1. Petar Bojovic, Slavko Gajin, "Testiranje i analiza funkcionalnosti internet domena Republike Srbije", YUINFO 2013, Kopaonik 2013. god. , (M63)

2. Bojan Mitrovic, Mirjana Devetakovi6, Slavko Gajin, Ljiljana Petrusevski, "Unapredenje AMRES e-Iearning sistema novim funkcionalnostima - aformat modul", YUINFO 2011, Kopaonik, 2011. god., (M63)

3. Mirjana Devetakovic, Slavko Gajin, Bojan Mitrovi6, "Portal Akademske rnreze Srbije za podrsku elektronskom ucenju", YUINFO 2010, Kopaonik, 2010. god., (M63)

4. S. Gajin, D. Pajin, D. Novakovi6, "Sistem za nadgledanje racunarske rnreze-NetltS", YUINFO 2006, Kopaonik, 6-10.3.2006. god., (M63)

5. Slavko Gajin, Pavle Vuletic, "Trendovi u razvoju i primeni racunarskih rnreza", Informatika 2004, Beograd, 2004. god., (M63)

6. Slavko Gajin, "Razvoj EDI aplikacija na Internetu", YU INFO '97, Brezovica, 04-07. april 1997. god., (M63)

7. Slavko Gajin, "Sigurnosni mehanizmi u protokolu za nadzor i upravljanje rnrezorn", naucno strucni skup Informacione tehnologije (IT '96), Zabljak, 11-15. mart 1996. god., (M63)

8. Slavko Gajin, "OSISS - Otvoren sistem implementacije sigurnosnih servisa", naucno strucni skup Informacione tehnologije (IT '96), Zabljak, 11-15. mart 1996. god., (M63)

9. Slavko Gajin, "Distribucija kljuceva u OSISS okruzenju", naucno strucni skup Informacione tehnologije (IT '96), Zabljak, 11-15. mart 1996. god., (M63)

10. Slavko Gajin, "OSISS - Otvoren sistem implementacije sigurnosnih servisa", YU INFO '96, Brezovica, 02-05. april 1996. god., (M63)

11. Slavko Gajin, "Model alternacije zaokreta kod resetke", YU INFO '95, Brezovica, 04- -07. april 1995 god., (M63)

12. Slavko Gajin, "Pobolisan]e algoritma rutiranja u cvrstospregnutim rnultiracunarskim rnrezama", XXXVIII konferencija ETRAN, Nis, 07-09. jun 1994. god., (M63)

Predavanja po pozivu i tutorijali na skupovirna nacionalnog znacaja:

1. Slavko Gajin, "Pristup i servisi Interneta", Naucno strucni skup INFORMATIKA 96, Beograd, 07.05.1996. god. , (M62)

2. Slavko Gajin, "Sigurnost i zastlta u racunarsklrn mrezarna", Strucni skup, Zastita podataka u racunarskirn rnrezarna i sistemima, Beograd, 1995. , (M62)

3. Slavko Gajin, "Neki aspekti sigurnosti UNIX operativnih sistema", Strucni skup, Zastita podataka u racunarsklrn rnrezarna i sistemima, Beograd, 1995., (M62)

Disertacije: 1. Slavko Gajin, "Analiza adaptivnosti modela zaokreta u cvrstospregnutim

multiracunarskirn mrezarna", magistarska teza, ETF, Beogard, 1999., (M72)

2. Slavko Gajin, "Opsti model determlnistickoq rutiranja u multiracunarskirn rnrezarna" doktorska teza, ETF, Beograd, 2007., (M71)

Strana 3

Projekti tehnoloskoq razvoja Ministarstva prosvete, nauke i tehnoloskoq razvoja:

1. "Opsti elementi i posebne primene zastite podataka u racunarskim sistemima i mrezarna", evidencioni broj S.1.02.05.0163, 1997.-2000. god.

2. "Projekat realizacije integralnog informacionog sistema i monitoringa racunarske rnreze'', evidencioni broj 1-253,2002.-2004. god.

3. "Razvoj kompjuterskih metoda i softvera za modeliranje i simulacije u oblasti opsteq i biomedicinskog inzenjeringa", 2005-2007. god.

4. .Prostornl, ekoloski, energetski i drustveni aspekti razvoja naselja i klimatske promene - medusobni uticaju", evidencioni broj 36035, 2011-2017. god.

Clanstvo u programskim odborima medunarodnih konferencija:

1. Member of Technical Program Committee of International Conference on Electronics, Computer and Computation ICECCO (www.icecco.org)

2. Member of Technical Program Committee of ICT Conference & Exhibition - INFOTECH (www.infotech.org.rs)

3. Member of Technical Program Committee of Smart E-Government International Conference and Exhibition (www.smartegov.rs)

4. Member of Technical Program Committee of Conference ICT INOVATIONS, Association for Information and Communication Technologies ICT-ACT (www.ict-act.org)

Clanstvo u medunarodnim i dornacirn strucnirn telima:

1. Nacionalni predstavnik u strucnom telu .e-lnfrastructure Reflection Group" (e-IRG), koje ima savetodavnu ulogu Evropske komisije u oblasti elektronskih infrastruktura

2. lnzenierska komora Srbije

Medunarodni projekti:

1. H2020 GN4-phase2 - "Multi-Gigabit European Research and Education Network and Associated Services", Grant Agreement No. 731122, 1.5.2016. - 31.12.201B., www.geant.org, clan tima

2. H2020 GN4 - "Multi-Gigabit European Research and Education Network and Associated Services", Grant Agreement No. 691567, 1.5.2015. - 30.B.2016, www.geant., clan tima

3. FP7 GN3plus - "Multi-Gigabit European Research and Education Network and Associated Services", 1.4.2013.-31.3.2015., www.geant.net. zamenik rukovodioca tima

4. FP7 GN3 - "Multi-Gigabit European Research and Education Network and Associated Services", 1.4.2009.-31.3.2013., www.geant.net. zamenik rukovodioca tima

5. FP7 SEERA-EI - "SouthEast European Research Area - e-Infrastructure", 1.4.2009.-31.3.2012., www.seera-eLeu, clan projektnog tima

6. TEMPUS ViCES - "Video Conferencing Educational Services", 144650-TEMPUS- 200B-IT-JPGR, 2009. - 2011. gOd., vices.marnet.net.mk, rukovodilac tima

7. FP6 SEEREN2 - "South-Eastern European Research & Education Network", oktobar 2006. - april200B. god., www.seeren.org, rukovodilac tima

Strana 4

8. FP6 SEEFIRE - South-East Europe Fibre Infrastructure for Research and Education", 2005.-2006. god., www.seefire.org, rukovodilac tima

9. FP6 SEEGRID -"South-Eastern European GRID-enabled elnfrastructure Developement", 1.5.2004. - 30.4.2006., www.see-grid.org, clan projektnog tima

10. FP6 SEEGRID2 - "South-Eastern European GRID-enabled elnfrastructure Developement 2", 2006-200B. god., www.see-grid.eu, clan projektnog tima

11. FP5 SEEREN - "South-Eastern European Research & Education Networking", 2002. - 2004. god., www.seeren.org, rukovodilac tima

12. ELISA -"E-Iearning for improving access to Infomation Society for SMEs in the SEE Area" (INTERREG III B CADSES project), 2005-200B. god., www.elisa-project.net. clan projektnog tima.

Tehnicko resenle - priznat softverski sistem na medunarodnom nivou:

1. ICmyNet.Media - softverski sistem za analizu multimedijalnog mreznoq saobracaja na bazi Cisco media net tehnologije, Soneco d.o.o., verifikovano od strane Cisco Systems kroz program CDN (Cisco Developer Network), (MB1)

lzvodacki, glavni i idejni projekat racunarskih rnreza:

1. .Projekat informacionog obezbedivanja i tajnosti podataka u racunarskoj mrezi RZZO", Republicki zavod za zdravstveno osiguranje, 200B. god. (rukovodilac projekta)

2. Radio televizija Srbije, "Projekat racunarske rnreze Radio televizija Srbije, lokacija Radio Beograd", Radio televizija Srbije, 2003. god (odgovorni projektant).

3. "Glavni projekat lokalne racunarske rnreze zgrade prlrodno-rnaternatickih fakulteta Univerziteta u Beogradu", 2002. god. (odgovorni projektant)

4. .Prerada Glavnog projekta racunarske rnreze RTS-a na lokaciji Aberdareva, Nova zgrada", Radio televizija Srbije, 2001-2002. god. (odgovorni projektant)

5. "lzvodacki projekat racunarske mreze poslovne zgrade NBS u Beogradu", Narodna banka Srbije, 2000. god. (odgovorni projektant)

6. "Idejni projekat racunarske rnreze poslovne zgrade NBS u Beogradu", Narodna banka Srbije, 2000. god. (clan projektnog tima)

7. "Glavni projekat lokalne racunarske rnreze poslovne zgrade PTT-a u Takovskoj 2", JP PTT Srbija, 1999. god. (odgovorni projektant)

B. Idejni projekat, Glavni projekti i Projekti izvedenih stanja lokalnih racunarskih rnreza Elektrodistribucije Beograd (1B lokacija), 1998.-2000. god (odgovorni projektant za 10 lokacija)

9. Tehnlcko resen]e racunarske mreze PLATNET, NBJ, 199B. god. (clan projektnog tima)

10. Glavni projekat Racunarske rnreze Sluzbe za zajednicke poslove saveznih organa (6 lokacija), 199B. god (odgovorni projektant)

11. Projekat "Povezivanje heterogene racunarske opreme u Racunarsku rnrezu Federacije", 1997. god.

12. "Glavni projekat Racunarske rnreze ICN Jugoslavija u Zemunu", ICN Jugoslavija, 1997. god. (odgovorni projektant na 3 projekta, clan projektno tima na 2 projekata)

13. "Glavni projekat racunarske mreze Filozofskog fakulteta u Beogradu", 1997. god. (odgovorni projektant)

Strana 5

14. "Idejni projekat Racunarske rnreze DP Valjaonica bakra Sevojno" , 1997. god. (clan projektnog tima)

15. Projekat .Povezivanie Akademske rnreze Jugoslavije sa Internetom", 1996. god. (clan projektnog tima)

16. .Projekat uvodenja Interneta u SRJ", 1999. god. (clan projektnog tima)

17. .Projekat kicme Racunarske rnreze Federacije", 1996. gad. (clan projektnog tima)

18. Idejni projekat racunarske rnreze Federacije, 1996. god. (clan projektnog tima)

19. "Glavni projekat Racunarske rnreze Radio-televizije Srbije", Radio-televizija Srbije, 1996. god. (clan projektnog tima)

20. "Idejni projekat Racunarske rnreze Vlade Republike Srbije", 1995. god. (clan projektnog tima)

Rukovodilac iii ucesce u softverskim projektima ETF-a:

1. Razvoj, implementacija i odrzavan]e informacionog sistema za obracun i naplatu elektricne energije, RTV pretplatu i pruzanje elektrodistributivnih usluga potrosacima, EDB, 2006-2013.

2. Maticna evidencija osiguranih lica, RFZO, 2002-danas

3. Elektronsko fakturisanje lekova i pomagala propisanih na teret sredstava Republickoq zavada za zdravstveno osiguranje, RFZO, 2002-danas

4. Elektronska evidencija propisanih i izdatih lekova pod specijalnim rezirnorn izdavanja, RFZO, 2002-danas

5. Zdravstveni informacioni sistem primarne zdravstvene zastite, 2005-danas

6. Razvoj, implementacija i odrzavanje Informacionog sistem za magacinsko, materijalno i pogonsko knjigovodstvo, NBS-ZIN, 2009-2015.

7. Razvoj, implementacija i odrZavanje informacionog sistema za pracenje proizvoda NBS-ZIN od posebnog interesa, NBS-ZIN, 2007-2015.

8. Podsistem za podrsku rada pisarnice Republickoq zavada za zdravstvena osiguranje, 2009-danas

9. Razvoj, implementacija i odrzavanje Opstinskoq informacionog sistema - OplS, 2000-danas (35 opstina)

10. Razvoj, implementacija i odrzavanje Softverskog sistema za evidenciju maticnih knjiga, 2004-danas (33 opstine)

11. Informacioni sistem i monitoring racunarskih rnreza - NetllS, 2002-2013.

12. Testiranje DNS damena, RNIDS, 2012. god.

Rukovodilac ostalih softverskih projekat:

1. NetVizura NetFlow Analyzer - softverski sistem za analizu mreznoq saobracaja, Soneco d.o.o.

2. NetVizura EventLog Analyzer - softverski sistem za analizu syslog i snmp trap poruka, Soneco d.o.o.

3. NetVizura DNS Analyzer - softverski sistem za testiranje DNS domena, Soneco d.o.o.

Priznanja i nagrade:

Strana 6

1. Plaketa Drustva za informatiku Srbije za izvanredan doprinos u razvoju informatike u 2002. godini (projekat NetlS, ReUB).

Strana 7

YHHBEP3MTET Y J>EOrp AW

CTY)l.eHTCKI1 rpr 1, 11000 Beorpan, Penytinuxa Cp6uja Ten.: 011 3207400; <DaKC: 011 2638912; E-mail: [email protected]

BEnE HA Y1IHI1X OEJIACTH TEXHWIKHX HAYKA

Eeorpan, 18.11.2013. rornrae 02 6poj: 61202-5356/2-13 JI,lJ;

Ha OCHOBY qJI. 65. CT. 2. 3aKOHa 0 BHCOIWM 06Pa30BaIhY (ItCJIY)K6eIIH rIIaCHHK PCIt, 6poj: 76/05, 100107-aYTeHTHqHO ryvaseae, 97/08,44/10,93/12 H 89/13), 'IJI. 47. CT. 5. ras. 1. CTaTYTa Ymmep3HTeTa y Beorpany ("fJIaCHHK YHHBep3HTeTa y Beorpany", 6poj 162111-rrpe'IHllioeHH TeKCT, 167112 H 173/13), 'III. 13. CT. 1. ITpaBHJIHHKa 0 BenHMa aaysnnx 06JIaCTl-I aa YHHBep3HTery y Beorpazty ("f'JIaCHHK YHHBep3HTeTa y Beorpany", 6poj 134/07, 150/09, 158/11, 164111 H 165111), qJI. 21. CT. 1. ras. 1. ITpaBHJIHHKa 0 HaqHHY H rrocryrncy cranarsa snan,a H sacnasaa.a pannor onnoca HaCTaBHHKa YHHBep3HTeTa Y Beorpany ("f'JIaCIII1K Ymmep3HTeTa y Eeorpany", 6poj 142/08, 150/09 H 160/11) H KPHTcpHjYMa sa cruuan,e asan.a HaCTaBIIHKa aa YHHBep3HTery y Eeorpany (lJfJIaCHHK YHl1Bep3HTeTa y Eeorpazry", 6poj: 140/08, 144/08, 160/11, 161/11 H 165/11), a na npennor H36opHor Rena EnCI<TpOTCXHJ1lIKOr cpaKynTcTa Ymmep3MTeTa y Beorpany, 6poj: 1455/4 on 19. cerrrevopa 2013. romrae, Bene aaysnnx 06JIaCTM TeXHH"'!KHX aayxa, na cennau», onpacanoj 18. HOBeM6pa 2013. rozmae, ,n:OHeJIO je

O):(JIYKY

EI1P A CE ):(p CJIaBIW I'ajnn, y saaa,e nouenra sa Pasyaapcxa TCXHHKa H HHcpopMaTHKa.

06pa3JIO'KeU,e

YHHBcp3HTeT y Beorpazty - EJIeI<TpOTcXHH"'!KH cpaKYJITeT jc ziana 19. jyna 2013. rO,[(I1He y JIHCTY "ITOCJIOBH", o6jaBHo KOHKYpC sa H360p y asarse nouenra, sa Y)KY naysny ofinacr: Paayaapcxa TeXHHKa M mlcpopMaTHKa, 3601' HCTCKa H360pHor nepnona.

H3BeIlITaj KOMHcHje aa npnnpeay H3BeiliTaja 0 npajasrseaan Kan,[(M,n:aTHMa CTaBJbeH je na )'BH,n: jasaocra naaa 26. aBrYCTa 2013. ronane, aa HHTepIleT crpann cpaKYJITCTa.

Ha OCHOBY npennora KOMHCHjc sa IIPHIIPCMY assenrraja 0 rrpl1jaBJbCHHM KaH,[(M,n:aTHMa, H36oPHO Bene EJICKTpOTeXHlf'lJCOr cpaKYJITCTa, Ira CC,n:HI1U.H onpxcanoj 19. cerrreutipa 2013. ronnae, ,[(OHeJIO je O,n:JIyKY 0 )'TBpljHBaIbY npennora na ce xannnzrar )J,p CJIaBKO fajHH, H3a6cpc Y asan,e noueara.

<l>aKYJITeT je nana 4. HOBeM6pa 2013. rO)J,HHe )J,OCTaBHO YmIBcp3HTeTY KOMIIJICTaH 3aXTeB sa H360p Y snarse na npOIIYICaHHM oopacnaaa.

YmIBep311TeT je KOMIIncTHY )J,oKYMeHTau.Hjy xojy je .uOCTaBHO <PaKynTcT CTaBl10 ua web CTpaHHU.Y YHHBep3HTcTa, naua 12. HOBcM6pa 2013. rOJ(HHC.

- 2-

Bene aay-max o6JIacTI1 TeXHW!KI1X nayxa, aa Ce,1l.HI1~11 O,1l.P)IcaHOj ,1l.aHa 18. HoseM6pa 2013. rO,1l.MHe, p33MaTPaJIO je 3aXTeB <DaKymem B YTBp,1l.BnO na KaH,[(M,[(aT I'ICnYH>aBa YCJIOBe nponacane qrraHOM 64. H 65. 3aKoHa 0 BMCOKOM ofipasoeaa.y 1'1 qJIaHOM 124. CTaTYTa YHMBep3HTeTa y Beorpazty, xao 11 YCJIOBe nponacane Kpl'ITepHjYMI1Ma sa CTI1~aIi>e ssarsa HaCTaBHHKa na YHl1Bep3l1TeTY y Beorpany, na je .D:OHeTa o.D:JIyKa icao y H3pe~H.

lIoCTaBI1TI1: - <DaKYJlTeTY (2), - cexperapy Beha (l), - ApxHBI1 YHHBep3HTeTa (1).

I1PE.nCE,lJ;AB4\JY;1Hl BEilA

Ilpotp . .D:P MHJroc~~ OrI-baHOBHO

I! r \~~L{j>-t~

YR' Jjn:>E'P311AlT'ET v rr.()'r'UA TIy , " "i',l,n, ,..I, JIlt", ,:.i " e, A'lU:, , , ,n 0l4.

EJIEKTPOTEXHJ1lJKM (])AKYJITiET

Te:l: +38.1 11 3248464, <llall:c: +381 ! I 32.t86iH

Ha ocnoay 'IJIaHa 24 eras 1, 27, 30 H 171 cran 1 3uKona 0 pall)' ("CJlY>K6eIUi ntaCHHK PC' 6poj 2412005); )LOHOCHM

' .. ~.'.:; ~/r:.)jlVti{/). (:;P5i~·SA >i';;vmEP3VtTET Y EEOfPAttl,y

tllAKYJlTEl P E ill E I-b E o 3ACHI1BAl-h Y P A,l.(HOr' 01(110CA

HMeHOf3(:lHH he BpruHTH nOCJIOHe y 3BtlFbY X.OIH!HTa Ha

pa-rynapcxy TexHln~y 1-1 IfH(I)OpM~rrnKY 0):1, 18.11.2013. ITi;l,HHC.

)].p Cnamco Fajnn sacnnna panun oruroc ua orrpeheuo paJ1Hl1lVI BPCMCHOM 0)( 18.11.2013. )10 17.11.2018. roznruc.

3apa]{3 no OBOM penren.y he ce 06pULlYUClBUTH OJl 18.1 L20B. ronuue.

06pa3JIO:tRCIbC

Y CKJIallY ca norpefiasta npoueca P[UW a ua ocnouy Onnyxe Bella Hay'ulIlx 05JH1CT'H TeXH,l1I·IKHX nayxa 02 6poj 61202-54 70i2-1 3 OIl 18.1 J .2013. rO).lHfIC. j.,lOHCTO

jc PCUJClbC xao Y)J,Hcn03HTHBY.

llPABHA nOYKA: Ilporus onor pCmCI:bU 3UnOCJICIUl MO)KC ).{a noxpeue cnop npen IUlll)lC)KHHM cynoM y pOKY 0,[1, 90 zuura O)~ nana )JOCTaBJbaIh<l.

- HMcHoBaHoM - OI~ceKY sa Mcpl1 - Kanposcxoj cnY)I\6H

ApXHBH:

X(OCTaBMTI:C

Documents

Univerzltet erne Gore ElEKT'ROTEHNICKI FAKlJlTE1' …...Univerzltet erne Gore v ElEKT'ROTEHNICKI FAKlJlTE1' ucc 81000 Podqorioa, OZ. Vasingtona bb, tel. t.R. 510-255-51, PIS: 02016702