Task Sch PPSO

Embed Size (px)

Citation preview

  • 8/8/2019 Task Sch PPSO

    1/32

    Parallel Job Submission In Grid Environment

    UsingParallel Particle Swarm Optimization

    Dr. G. Sudha Sadhasivam

    Asst. Professor

    Dept. of CSE.PSG College Of Technology.

    D. Komagal Meenakshi (07MW05)

    PSG College Of Technology.

  • 8/8/2019 Task Sch PPSO

    2/32

    Outline Scheduling in Grid.

    Problem Statement Need For Job Grouping in Scheduling

    PreviousWork Done in Job Grouping

    Proposed System

    Trust Based Filtering of jobs

    Particle Swarm Optimization Parallel PSO

    Model for PPSO

    Dynamic jobs

    Results

    Conclusion and Future work Bibliography

  • 8/8/2019 Task Sch PPSO

    3/32

    Scheduling in Grid.

    Grid computing is a high performance computing

    environment to solve large scale computational demands.

    Task scheduling is a fundamental issue in achieving

    high performance in grid computing systems.

    Reason: Large numbers of tasks are computed on the

    geographically distributed resources, a reasonable

    scheduling algorithm must be adopted order to minimizejob completion time with uniform load distribution.

  • 8/8/2019 Task Sch PPSO

    4/32

    Need

    An unorganized deployment of grid applications with a

    large amount of fine-grain jobs

    Leads to

    communication overhead dominate the overall processingtime

    Low computation-communication ratio.

    Results

  • 8/8/2019 Task Sch PPSO

    5/32

    Need For Job Grouping in Scheduling

    Efficient job grouping-based scheduling system is required.

    A Grid Scheduler should

    Reduce the total transmission of user jobs to/from the

    resources.

    Reduce the overhead processing time of the jobs at the

    resources.

  • 8/8/2019 Task Sch PPSO

    6/32

    Job Grouping

    Dynamically

    assemble

    Transmit

    Grid resources

    job groups [ coarse grained ]

    Jobs of an application [ fine grained ]

  • 8/8/2019 Task Sch PPSO

    7/32

    PreviousWork Done in Job Grouping

    Comparison of Scheduling algorithms with and without jobgrouping.

    In the context of DAG scheduling, grouping of jobs into clusters

    to reduce inter-job communication.

    Job Grouping strategy, adaptive to run time environment

    Job Grouping with PSO.

  • 8/8/2019 Task Sch PPSO

    8/32

    Proposed System A noveljob grouping method using Parallel PSO

    To reduce the communication overhead.

    Enhance the speed of completion of processes.

    Improve resource utilization.

    Improve parallel efficiency.

    Uses PPSO to select the resources to minimize the make span.

    Trust level and dynamism of jobs is considered

    Tool Used - Gridsim-4.2-beta.

  • 8/8/2019 Task Sch PPSO

    9/32

    The Project aims at

    Job Grouping based on trust Using PPSO

    Parallel Job Submission

    Enhancing Computation-communication Ratio

    Reducing The Overall Processing Time Of Jobs Using

    Parallelization

    Improving Resource Utilization In The Grid Environment.

    Trust based job filtering

    Dynamic job submission

  • 8/8/2019 Task Sch PPSO

    10/32

    Dynamically assemble

    Using PPSO

    Transmit to

    Grid resources

    job groups [ coarse grained ]

    Filtered Jobs of an application [ fine grain] based on Trust

    Grid resources Grid resources

    In Parallel

    1. Job Grouping

    J b G i

  • 8/8/2019 Task Sch PPSO

    11/32

    Total number of jobs

    Average MI rate of job

    MI deviation Percentage

    Overhead processing time

    Granularity time

    Grid Resource

    Grid resource 0

    Grid resource 1

    Grid resource N

    Grid Resource File

    User Input

    GridletsGrid resources characteristics

    Gridlet MI Resource MIPS Granularity time

    Total MIPS

    Grid resource 0

    Gridlet group 0

    Grid resource 1

    Gridlet group 1

    Grid resource 2

    Gridlet group 2

    Gridlet groups Resource IDs

    ..

    Gridlet Scheduler

    (1)

    (3)

    (4)

    (5)

    (6)

    (7)(2)

    Trust level

    In parallel

    Filter jobs

    based ontrust

    Job Grouping

  • 8/8/2019 Task Sch PPSO

    12/32

    2. Trust Based Filtering of jobs

    The Grid Information Service GIS gives the information

    about all the trust level of the resources .

    The user submits the jobs with different trust values.

    From this, the jobs that have trust values greater thanthe resource's trust value are filtered out.

    Trust aware resource management and scheduling offerQuality of Service at application layer in grid

    environment.

  • 8/8/2019 Task Sch PPSO

    13/32

    3. Particle Swarm Optimization If large numbers of tasks are computed on the

    geographically distributed resources, a reasonable

    scheduling approach must be adopted in order to

    get the minimum completion time.

    Task scheduling is a NP-Complete problem

    Heuristic optimization algorithm can be used tosolve NP-complete problems.

  • 8/8/2019 Task Sch PPSO

    14/32

    Particle Swarm Optimization (PSO) is an evolutionaryoptimization technique inspired by nature.

    It simulates the process of a swarm of birds preying.

    Its global searching ability can be used for neuralnetwork training, control system analysis and design,

    structural optimization.

    It also has fewer algorithm parameters than geneticalgorithm.

    PSO algorithm works well on most global optimalproblems.

  • 8/8/2019 Task Sch PPSO

    15/32

    PSO Concept

    A swarm intelligence based algorithm finds a solution toan optimization problem in a search space.

    Proposed solution exists in the form of a fitness function.

    The swarm is typically modeled by particles inmultidimensional space that have a position and avelocity.

    A Particle is a candidate solution in the population andrepresents a task.

  • 8/8/2019 Task Sch PPSO

    16/32

    Particles fly through hyperspace .

    An iterative process to improve candidate solutions is set in motion.The particles iteratively evaluate the fitness of the candidatesolutions.

    Particles posses two essential reasoning capabilities Memory of their own best position and

    knowledge of the global best of the swarm.

    As the swarm iterates, the fitness of the global best solutionimproves.

    All particles being influenced by the global best eventually approachthe global best. This phenomenon is called 'convergence'.

  • 8/8/2019 Task Sch PPSO

    17/32

    PSO Algorithm Initialize parameters

    Initialize population randomly

    Initialize each particle position vector and velocity vector

    Do {

    Update each particles velocity and position;

    Find a permutation according to the updated each particles position; Evaluate each particle and update the personal best and the global best;

    Apply the local search;

    } While (!Stop criterion)

  • 8/8/2019 Task Sch PPSO

    18/32

  • 8/8/2019 Task Sch PPSO

    19/32

    Parallel PSO

    Recent advances in computer and network technologies led to parallel optimizationalgorithms.

    Parallel PSO (parallel implementation of stochastic optimization alg)

  • 8/8/2019 Task Sch PPSO

    20/32

    Parallel PSO design

    Intialize

    f(x) f(x) f(x)

    Check Convergence

    Update

    # of particles

    #

    ofiteration

    s

  • 8/8/2019 Task Sch PPSO

    21/32

    Model for PPSO

    Master

    Slave Slave Slave

    SEND GLOBAL VALUERECEIVE INDIVIDUAL

    VALUE

  • 8/8/2019 Task Sch PPSO

    22/32

    Gridlet Grouping

    Scheduler

    Trust based

    filtered Gridletlist

    Resource list

    Call PPSO to assign Gridlet To Resources

    Create new grouped Gridlet

    With length= Total length

    Assign to resources

  • 8/8/2019 Task Sch PPSO

    23/32

    4. Dynamic jobs

    Dynamic submission of jobs is considered. User can submit jobs when other jobs are being

    processed.

    The unused MIPS rating of the resources can beutilized in a efficient way such that grouping isdone by considering the unused MIPS as totalMIPS and the jobs are processed.

    Then Parallel Submission of grouped Gridlets toresources is done

  • 8/8/2019 Task Sch PPSO

    24/32

  • 8/8/2019 Task Sch PPSO

    25/32

    Simulation Time for Job Grouping

    using PSO vs. Parallel PSO

    90

    100

    110

    120

    130

    140

    150

    20 40 60 80

    No of Gridlets

    SimulationTim

    PPSO

    PSO

  • 8/8/2019 Task Sch PPSO

    26/32

    Total number of processed gridlets for

    different granularity time and resources

    0

    20

    40

    60

    80

    100

    R 1 R 1-R 2 R 1-R 3 R 1-R 4 R 1-R 5

    Resources

    NoofGridletscom

    pletedingran

    time

    10

    20

    30

    40

    50

    d d b

  • 8/8/2019 Task Sch PPSO

    27/32

    Load at resources during job grouping

    with PPSO

    0

    100

    200

    300400

    500

    600

    700

    800

    900

    1000

    R1 R2 R3 R4 R5

    Resources

    loa

    50 gridlets

    60 gridlets

    70 gridlets

    80 gridlets90 gridlets

    Diff i b i i i f idl

  • 8/8/2019 Task Sch PPSO

    28/32

    Difference in submission time of gridlets

    with PSO and PPSO

  • 8/8/2019 Task Sch PPSO

    29/32

    Add load balancing feature graph here

  • 8/8/2019 Task Sch PPSO

    30/32

  • 8/8/2019 Task Sch PPSO

    31/32

    Future Work

    Future work would involve developing a more

    comprehensive job grouping-based scheduling system that

    takes into account QoS (Quality of Service) requirements of

    each user job before performing the grouping method.

    Resource utilization can be done according to the capacity

    of the resource.

  • 8/8/2019 Task Sch PPSO

    32/32

    THANK YOU