A MULTICHANNEL COMPUTER NETWORK

WITH LOCAL AND GLOBAL TR ANSCEI VING MEDIA

By

WONG PO CHOI

黃 寶 財

A MASTER THESIS SUBMITTED IN PARTIAL

FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF

MASTER OF PHILOSOPHY

IN

THE DEPARTMENT OF ELECTRONICS

THE CHINESE UNIVERSITY OF HONG KONG

HONG KONG

MAY 1985

ABSTRACT

A new local area network LAN topology is proposed. I t

consists of a multichannel global transceiving medium and a group

of local transceiving channels. A station uses the local channel

to communicate with other stations on the same local channel and

uses one of the global channels to communicate with stations on

the other local channels-

A new reservation protocol Contention-based Look-Ahead

Reservation protocol for Multichannel Communications (CLAR-MC)

is proposed for this topology. Various properties of this

protocol are studied and simulation results on message delays and

throughput are obtained. A two dimensional Markov chain model is

built or the network protocol. The performance obtained from the

Narkovian analysis agrees very closely to that obtained by

simulation.

The proposed network can offer an integrated service to

different app l icr-it jons. I t can support message switching, packet

switching as well. as fa: t circuit switching. It can provide

services such as message data transmission, two party telephone

calls, multiparty telephone conferencing, fascimile as well as

providing circuits with customized data rate.

A CI NO WL E DG E M E NT S

I would like to express my deepest appreciation to my

research supervisor, Dr. T.S. YUM, for his continuous advice,

guidance, encouragement and support.

TABLE OF CONTENTS

1. INTRODUCTION

2. A NEW COMPUTER NETWORK TOPOLOGY

2.1 Topology

2.2 Network Access Protocol

3. PRELIMINARY PERFORMANCE EVALUATION

3.1 Constraint on average message size

3.2 Maximum Throughput

3.3 Average Delay

4. APPLICATIONS ON VOICE TRAFFIC AND CONFERENCING

4.1 General Discussion

4.2 Voice Traffic in CLAR_MC

4.3 Preliminary Performance evaluatioin

5. ANALYTICAL MODEL

5.1 Assumptions

5.2 The Network Model

5.3 Calculations

5.4 Numerical Results

6. SIMULATION RESULTS

7. CONCLUSIONS

REFERENCES

APPENDIX : Simulation Program in GPSS

1

5

5

7

13

13

15

16

18

18

22

30

34

34

36

40

42

43

46

47

CHAPTER 1

INTRODUCTION

A computer network has a number of computer systems,

terminals, peripheral devices connected together by a

communication medium. Through the network, computers can

communicate with each other, share each other's information, and

perform tasks together. Although networks, of various types have

been implemented, a definitive design is still far from the

rp A 1 m

The Local Area Network (LAN) is a special type of computer

network with all the stations (computers, terminals, and other

equipments) located in a small geographic area, e.g. within a

radius of two miles. Typically, such a network is owned by a

single organization and so tends to be specialized in

applications. For example, a business firm may use a LAN to

enhance office automation. Documents can circulate among

departments through the network and comments and responses can be

returned almost immediately. This can greatly improve the

efficiency of an organisation. Also expensive peripheral devices

like typesetters, printers, disk and backup tape storages,

plotters etc. can be shared by different computer stations in

each department, resulting in lower equipment cost and better

i] f i 1 l -t: f i on of o y nn 3 i f pr i I i f i p p-_

2

Many different types of LANs are proposed and implemented in

recent years. Among them, the Carrier Sense Multiple Access

(CSMA) Bus type computer network and the Ring type network have

received the most attention. The performance of the major types

of local computer networks are analyzed and compared in detail in

a paper by Stallings [1]. Bowerman [2] also made a broad

comparison of several specific local computer networks being

implemented.

Among the local computer networks compared, the CSMA Bus

type network (e. g. the famous Ethernet) has the following

advantages

1) It gives the smallest average delay when the

communication channel is lightly loaded.

2) The communication protocol is easy to implement.

3) The ability to broadcast packets.

4) High reliability due to passive interface.

5) Extremely flexible, i.e. stations can be added or deleted

easily without interrupting the network service.

3

However, CSMA type networks also have some drawbacks. First

of a l l, if the workload of the communication channel surges

temporarily, packet collisions on the channel will increase. All

stations are then busy at retransmitting their data packets that

are bound to be collided again with those from other stations. In

such a case the channel throughput will drop to very low and the

delay of successfully transmitted packets will be very large. The

second problem is that the performance of the CSMA type network

depends on the cable length of the network, the packet size and

the total number of active stations in the network. Thus if the

data packet size is small, the cable is long and the number of

stations is large( which are common requirements), the network

performance will be poor[5]. For this reason, the CSMA type

networks are usually smaller in size compared to other types of

networks in order to achieve acceptable performance.

Other types of computer networks such as Token Ring, Token

Bus, Contention Ring, Buffer Insertion Ring etc., all have their

advantages and disadvantages. Generally it is not possible to

tell which network is the best without specifying the application

areas, the hardware and software costs, the performance required,

and the kinds of computers and equipment to be linked on.

M，W ，，]〇 " , n u l U( ) ! 1 1 n , ' 1 1 -1 I [ > r 〇|K) .S(Mi ,, „ m 1 | i c ! 1 < m n e f (；r , M A -

tYpe c〇mPUte r network. They found that by dividing the channel

into M subchamiels (may be implemented by M cables of smaller

oandwidth ) , the throughpul： is increased. However, the protocol

used on Lhe network would require M receivers for each station

and the station has to deLect status of all [：he su bcfiannels

for. finding an idle channel for tra nsm i ss ion „ This means that

complicated and costly tiransceiver is ne€)r]ed for each station.

I^'rrorm.uK-o ( i (i< 11. i—()n to l:he incre.u^o of； cal.Lo lenqth

(using the CSMA protocols) still exists, which limits the size of

the network. rl' h e re s u 11.1 n cj n e t w o rk , therefore 7 is neither well

suited for low cost microcomputer communications nor for large

⑴ k 」 l〇n(j disUmce pdcket: trcuismissiori.

To combine trie advantages of the CSMA protocol and

multichannel top eulogy , we propose, in the following, a new LAN

topology "Multichannel Computer Network witfi Local and Global r ranscei viny Media'1 and a new communication protocol to be used

on 1 * the "Contention-based Look-Ahead Reservation protocol for M u It .i c h a n n e ]. Com mu n ic<i t Ion s ( CL AR-MC ) 11 .

It is shown that: the proposed network topology has many

advantaye^ compared with other network topologies. It can support

0 o t h packet t ran sm i. .s g i on <-i n(1 voice c i r c u i (: reservation. It is

we 丄」 dpp J—j. !:〇 ‘i i ol: 〇[； services such a s me ssaqe

Lr arismission , vo_i.ee calls, tie 1 e p hone con f e r e n c i n g f fascimile,

1 〇w — spe g ci virtual f： i rc u ;L t reservation e tc .

4

CHAPTER 2

A NEW COMPUTER NETWORK TOPOLOGY

2.I Topology -

Let the network be consisted of a multichannel global

transceivi n g medium and a group of .Local transceiving channels

(figure 2.1). All local and global channels are of the same bit

late ln order 1:0 ,nlnimize the transceiver c o f n p 1 e x i t i e s. Each station uses tlie local channel to cominu n i( • a t e with other devices

on the same local channels If a station wants to communicate with

a dlf^erent station on another local channel； it will do so via

one of the global channels^ To offer a diverse grade of services

and to control congestion if occurs, prot(3cols can be designed so

33 t o restrict certain stations from accessing certain global

channels. An example of network interface for the multichannel LAN is shown in figure 2,2.

There are a number of advantages in using this kind of network topo.1 oqy ；

1 Hlgh reliability : No station can monopolize the entire

communication channel. Even if a malfunction station keeps on

sending packets, i t will not flood the entire network with

traffic. Passive in i;er faces are used in both local and global channels to enhance re1iahi1ity。

2) I i i y Fi f 1 e x i b i J. i t： y a m (i in 〇 < j u 1 a r i t y : Stations and subnetworks

(or locri 1 mf：iy be added and deleted easily.

5

M

GLOBAL

CHANNELS

1

2

M-1

M

LOCAL

CHANNELS

DATA

CHANNELS

SIGNALLING

CHANNEL

Stations that can acess both local and global channels

Stations that can access only local channels.

Figure 2.1: Network topology of the multichannel

computer network with local and global

transceiving media

ylj»i .n

U OCTROM

MicrocornoutersTerminals,Par!r.ririTi1r-

MICROPROCESSOR

DMA

r.nnr rn1 1 PT-

Serial

Porti rNT-hv~r I I r

ITransceiver

•Control

TiPCi r

Transceivei

Transceivei MUX

Surfer Memory

Fi p-n rp??• Npfwnrk Tnfprfpp fnr Mnl ti phannpl T A N

SIGNALLINGCHANNEL

r at A4.Ti rtLki.o L, 0

6

3) Higher throughput: Local channels can simultaneously

transport "local" traffic on them. The effective traffic allowed

on the network is therefore increased.

4) Adaptability and Expandability: An organisation can choose

the number of local and global channels in the network to adapt

to its own need. I f it has only microcomputers to be linked, it

can implement just the local channel. The global channel and

other local channels can be built t at a later time when network

expansion is called for.

5) Larger geographic coverage: By using many smaller bandwidth

channels as the global transceiving medium the throughput

performance is less dependent on cable length. Also new

protocols can be designed to make the throughput even less cable

length dependent than the CSMA protocols, Moreover, since the

local channels can take up a proportion of the network traffic,

we expect the performance of our proposed network to be better

than the multichannel. CSMA network where all channels are

identical. Thus the network in principle can cover a larger

geographic area.

6) Cost-effectiveness: Since the subnetwork (the local

channel and its stations) size is relatively small, simple

protocols may be used on it to reduce the transceiver cost. Also,

single-user microcomputers need only be connected to one or two

global channels to reduce the cost of interface. For larger

computers and expensive equipments, it then justifies the use of

more expensive hardware for more efficient global communications.

7

7) Better, Security and Privacy: There is no way for stations

on one local channel to hear "conversations" on other local

channels.

2.2 Network Access Protocol

There are two access protocols to be considered, the local

channel access protocol and the global channels access protocol.

Since a local channel is shorter and is connecting only a

fraction of the total stations, simple CSMA/CD [1,5,17] protocols

can be used with satisfactory performance at very low

implementation cost. The procedures and performance of CSMA/CD

protocols are well understood and so we do not describe it in

detail here.

In this thesis we proposed a new network access protocol for

the global channels that would require less complicated

transceivers than the M-CSMA[3,4], and give higher throughput and

acceptable delay at high load. The protocol is termed

"Contention-based Look-Ahead Reservation protocol for

Multichannel Communications (CLAR-MC)". For an M channel network,

one channel is dedicated for signalling which includes functions

such as channel reservation (transmission request),

acknowledgements, channel re leasing, system set up etc. The

remaining M-l channels are for data transmission and are shared

by all stations through reservation on the signalling channel.

We choose a modified Slotted ALOHA protocol (see figure 2.4

and 2.6) for contention on the signalling channel. This protocol

requires a very simple hardware implementation and its

performance is well understood. But more importantly its

throughput performance is independent of the cable length and

message size[5]. Hence it is particularly suitable for networks

with large geographic coverage.

Let Active" stations be those which are ready for message

reception; and "Ready" stations be those with message ready for t r a n sin i ss i on. A ready station must also be an active station.

Each station Keeps a table recording the current information on global channels. A typical table looks as follows:

- - - - - - - - - - — — ^ - — ■— — — 一 一 .一 ，一 一

Channel Busy/Ready Transmit Receive Number /Idle Station Station___ _______________________ ____________ ________________

1 B 1 1002 R x x3 R 3 2 9

* * • .

° 筆 . •

9 I x x_ 丨 —*■ -- "■-撼n ~~ — II --mm. -II. . . _ - - r - 1

Whe re

Busy/Ready/Idle : Channel status, "Busy" means it is being used

for transmission and is not available for reservation through contention, "Ready" ni e a n s it is opened for contention b y

stations. A station succeeded in reserving the "Ready" channe 1 has to wait for the end of the previous transmission before it can start trans【nissior匕 "Idle" means that no station is using the channel and is ready for immediate transmiwssion.

Thus in the above table, for example, channel 1 is being used by stations 1 and 100, channel 2 is ready to accept reservations and channa 1 9 is idle.

m Channel Status Table

9

2.2.2 Station power-up procedure

When a station is power up, it can initialize its table by

broadcasting a request information packet through the signalling

channel. The station which is transmitting at the lowest numbered

channel answers the request by sending a packet containing the

table information to the star up station. It no such table is

teceived within ceetain time interval. say 10 slots,and all

the channels are detected to be idle, the startup station simply

assumes that there are no existing active stations using the data

channels. The transmitting probability p is set to 1 and the

status of a 11 channe1s a re set to I( id 1 e).

2.2.3 Message Transmission Procedure

When an active station becomes ready to transmit a message, it

follows the steps below:

Step 1: Look up the Channel Status Table to see if the

destination station is busy in transmitting or receiving

messages. If so, the source station will wait until the

destination station is ready to accept messages.

Step 2: Select an Idle channel for data transmission. If none

exists, choose a Ready channel at random.

Step 3: Transmit a reservation packet (see fig.2.3) through the

signalling channel to reserve the chosen channel. The

contentien nlgorithm is slown in rigure 2.4 If a

station succeeds in transmitting a reservation packet,

Channel 1

Channel 2

Lookahead Interval

Packet from station A

Packet from

station B

Activities

on

Channel k

A E ACK

Activities

on

SignallingChannelt

Station A transmits a release packet A? through the

signalling channel at the beginning of the lookahead reservation

interval. Station B succeeds in transmimitting a reservation

packet B~V after 1 slot of collision with other reservation

packets. After a reservation acknowledgement packet is received

fraii the destination station, station B then waits and starts

transmitting its message as soon as station A completes its

transmission.

Figure 2.B: Channel Reservation Timing Diagram

Conten t LotProcedu re

Wait: forslot start

WaL t; ReadyChanne1

Available

Y

Iransmit

Prob.= p

1- pDelay transmit

i Ion i tort; h e c h.

Transmit

reser.pack

Col 1 is ion?

. YCollisior

N

SuccessSlot?

V(

rX

p- p2

H

p- p2

Next Slot

Sucees fReserv.

OUT

Figure 2.4: Flowchart for signalling channel contention

Uie i m me (i i < \ t e 1 y f o 1. L ow i n g s l o t 〕‘. s reserved for the

(]g :v 匕 i iia t i. on s t:a t on to reply with a "R e so r va t i. on

Acknowledgement *f or "RACK" packet. All other stations

will re(-rain from t; ra ri sin i ss i on in t.fial:; slot.

Step 4 : After a RACK pricket is received from the destination

staL ion , message transmission inay begin on the "Ready"

(：!lfUlno1' ：；(j 1 or： !：〇(] (1 n〇〇n iL (joe ifJlo, A ] I

stations then change the status of that channel tonBusy " •

Step 5 : When Uiero flre !)its remain ing to he sent:, the source

station broadcasts a "Channel-Release" packet through

the signal1ing channel. All stations then change the

staL u .s 〇 l t i l a t data channel to M R e a d y " . a ”Ready '，

channel therefore may not: necessary be idle but it must

oecorne idle v/it/iin the t ra n .sin i ss i oti time of b bits. We

cie f i ii g U 丨e 1 s t: 1) l)i l;s t ra nsm i .ss 1 on interval a s the

"Look a he. Hi Res- rva Lion Interval,r „ This M lookahead"

mechanism is deliberately introduced so that the

transmission of the present message on the data channel

and the scheduling of: the next message on the same

channel are done sirnu Itaneously. The next message can be

t ra n sm 1 L (： oci 〇 s soon a s t fie f)i;e se n I: me ssaqe l: ransm i. ss i on is finished.

； 6 ： Afli^r r^coivLng the entire message, the destination

; :;L<iLion v/ill ,K'knowIedge the source station with a "ACK"

"UA(_K" (i\h. i \j <' k n()w 1 (、（i(|(、m(:、n ） p<i(，k(、t.

11.

F i g 2 . 5 is a flowchart summarizing the procedures for l-r.msini tti nc] a me5>：Mc]G. Figure 2.4 shown the contention algorithm

f〇L t:ho signal l i n g channel to tranmit the reservation, the

release and the RACK packets. Figure 2,6 shows the algorithm for

an aCtlve statlion to update U ig transmitting probability p. Figures 2.7 and 2,8 shov./ the typical formats for signalling

packets and data messages respectively. Figure 2.3 shows the

assignment of a data channel to a station after reservation at

tne signalling channel. The channel release packet is assigned a

Higher Priori t:y U w n hhe reservation packet by assigning it a

larger probability of l: ra nsm .1 ss i on This ensures that: channel

release information is quickly propagated to all stations so that

the data channel may be used again for message transmission„

MessageTransmission

Procedure

ReadyIdleChanne1

Available?WAIT

Con tention

Success?

Wait forRACK

Wait for assignedReady channel to

become idle

Transm i. tDa ta

At the beginningof hookahoad

In t:erva I

Contention

for signallingchannel

Transmit; channe 1

Re 1 e a s e p a c k e t:

Transmi. tDa ta

Ready fornext message

[•i'Mirt; 2.5 :Flowchart for the transmission of

'' ;t in( shy a station

Update pprocedure

Ready Ch.Ex ist?

NOUT

v1

Monitorthe reser.

C! in nne1

Co 11 is ion Success J d I e

p- p2 p-p p=p2

pl Y

p= 1N

OUT

Fi_gutg 2.6; blowchar t .for acf iv0 s ta tion to

update, the transmit probability p

Station 〇 n ch fnnel destination Error。 。 belect Station Check

Address Bits

a) Channel Reservation Packet (for message transmission)

S t ^ ^ o n n i y v Signalling ErrorL d - e s s X Information CheckAOu,css (dats) Bits

D) Signalling Packets for message transmission(include RACK, Release, other information packet etc.)

, 〇 y y Signalling Error, a L -on 1 0 X X Information CheckAdclress (voice) Bits

c) Signalling Packets for voice call

(include call setup, call ACK, talkspurt-circuit Reserv call r e l e a s e ) ，

p n ~ ~ n ------------------- //-------- -----------Source Multidestination Error

^ cltL〇n 1 1 x x broadcast/conference Checkaddress call information Bits

---------------------------------- _~ ~ ~ J~ ^ --------------------------------------------------------------/ / --------------------------- ----- -------------------------------d) Special signalling packets for multidestination call

It contains addresses of all the destination stations Lach destination adclress has one bit added to indicate wnetner it needs to send back an RACK or not. A special Broadcast/Conference Code is assigned.

Figure 2.7: Formats of Signalling； Packets

Destination

Address/ DATA CRCBroadcast MESSAGE ChecksumCode

a) Data Message: Variable Length Allowed

Destination

DigitizedAddress/ CRCConference Voice/Data ChecksumCode

b) First voice packet of a. talkspurt

DestlnaLion

Preempt DigitizedAddress/ CRCHeader Conference Voice/Data Checksum

Code

c) Subsequent voice packets

Figure 2.8: Format of Data Messages, First Voice Packet and

Subsequent voice packets

CHAPTER 3

PRELIMINARY PERFORMANCE EVALUATION

We first define the following:

M: Number of data channels

C: Capacity of each channel (Bitssec)

N: Number of active stations

Message arrival rate per station

(Assumed equal for all stations)

3: Average message size in number of bits

P: Size of small signalling packets in number of bits.

(Small signalling packetsincludereservation

packets, release packets, ACK packets etc).

The total amount of traffic on the M-l data channels is

N R K i f~ c? c:

3.1 Constraint on Average Message Size

There are three types of traffic on the signalling channel:

reservation packets, reservation acknowledgement packets and

release packets. As explained before, there is always a reserved

slot for the RACK packet. The number of slots required for RACK

for eack message transmission is always one. Let n1 be the

average number of contention slots (P bits per slot) required for

a successful transmission of a reservation packet. Let n2 be that

required for a successful transmission of a release packet (See

F i g. 3 .1). W e h a v e n 2= n1 since a higher probability of

transmission at a contention slot is assigned to the release

packet. The average number of signalling channel slots needed for

Rel Req R a c k Rel Req R a c k

n2 nl Idle n2 nl-

nl reservation packet contention

interval in slots.

n2 release packet contention

interval in slots.

Figure 3.1: Packet transmission on the Signalling Channel

a me ssaqe transmission therefore is H ence the total

traffic on the signalling channe1 is bps.

The traffic surely cannot exceed the channel capacity. Hence

we require:

for Data Channe1s

for Signalling Channel

In order to fully utilize the data channels we take

(M-1 )C. Upon subs titut i ng into the second inequality we obtain

the constraint on the message size as or the average

message size in slots must be 1arger than

The value of n depe n d s on the number of active stations N,

the traffic rate the number of data channels M -1 and the

contention algorithm used at the signalling channel. From

simulation, n is found to be between 3 and 5.6 depending on

network load.

3.2. Maxim um T h ro ugh p u t

From previous discussions we know that it the average

message size exceeds (M-l)nP bits, all M-l data channels are

fully utilized. Hence the maximum throughput is

S (max]

M- 1

and is independent of number of stations, loading and cable

1 e n a t h.

Simulation results show that this maxmimum throughput value

is obtained at an offered load of only about one. This means that

very high throughput may be obtained if the average size of

messages is sufficiently large in size, say 50 times the

contention slot size for M-10. Since a data channel once reserved

can be used to transmit the entire message, the average message

size is effectively larger compared with that of the single cable

network, in which case messages must be broken into smaller

packets, adding headers and transmit through contention one by

one. This is essentially fast~circu.it switching on a

multichannel LAN.

3.3. Average Delay-

Here, the data channel is assumed to be error-free. The

message delay is the sum of queueing delay for a ready channel,

reservation delay, delay in waiting for the reserved channel to

become Idle and the message tranmission delay. The reservation

delay, moreover, is composed of the contention delay, the

reservation packet transmission time and the RACK transmission

t ime.

Under light loading condition, say S 0.3, a lot of idle

channels are available. Therefore

1) Queueing delay for Ready Channel= 0

2) Average Contention Delay

by assigning nl=l assuming there is

very little collision at low load.

3) Waiting Delay after reservation= 0

4) Average Transmission time- BC

Therefore the average total delay D is (2P+ B)C.

An equivalent channel with combined channel capacity MC will

have average message transmission time T'equals to BMC. The

average total message delay D normalized with respect to T' is:

Since 2PB 1, the de1 ay on the M-channe1 network und(

light loading condition is about M t i m e s the delay on a sing]

channel network. In other words, the message transmission time:

the dominating delay.

17

Under medium to heavy traff is conditions, say 0.3 S 0.8,

the queueing delay is small compared to the reservation delay

plus the message tranmission time, and can be neglected. A

station after reserving a ready channel has to wait for that

channel to become idle. If the lookahead interval is sufficiently

large, channel reservation is done before the end of the

lookahead interval. Hence, the maximum waiting delay is equal to

the lookahead interval b/C. Therefore the average total delay D

is (3+b)/C and the normalized average delay D' is M(l+b/3), or

larrier than M but less than 2M.

Under very heavy traffic condition, S 0.8, the

queueing delay dominates and the delay will blow up as in any

other resource sharing systems.

The selection of lookahead interval size is subject to two

contradicting factors. If the interval is too short, contention

on the signalling channel may go beyond the lookahead interval,

with the throughput decreased. I f the interval is too long,

unnecessary delay in waiting for the end of the interval may

occur when some other channels 'become idle in this interval.

Let LI be the size of lookahead interval in number of slots.

Simulation result shows that 5 LI 30 slots all give

comparable performance. A value of T I equals to 10 slots is

chosen.

CHAPTER 4

APPLICATIONS ON VOICE TRAFFIC AMD CONFERENCING

4 •丄 General Discussion

Since the introduction of computer networks, the possibility

Ox. integrating d.i.]：i.erent types of traffic onto a network lias been

investiga ted. In p a r t icu1ar , the integration of voice .nul dnl：a

trdffiC extensively studied [6,7,8]. John G. Gruber[8]

(]a V〇 11 丨 rovl〇w ()[ erUmnced circuiL, park e t: , and hybrid

switching techniques applyinq to voice traffic, lie then d is m s sod

the delay performance of these techniques. The network topology

concerned is a general mesh type or structured hierarchical type that: are used f：〇r packet switching.

Integration of voice and data has also been proposed on a

bus type local area network using protocols that are

modifications of CSMA/CD [9,10,11] . However as pointed out in

[1〇], due to the special nature of voice traffic, ordinary

CSMA/CD protocol.^ are n〇[： well suited for such integration.

Speech signals are prepared for digital transmission either

by speech vocoders or waveform coders. Vocoders analyze

iucr： 〇 i ve i xocj rv^tls ( frames ) of U)e input speech.

Typically, the duration of these frames i:3 between 10-25

ms[9][ll」. The exact nature of the framing varies accordinq to

the coding algorithms used. Each frame is represented (〇r

encoded ) by a .small number of bits and constitutes one "parcel"!*■

of speech. Voice parcels are generated periodically during

talkspurts only. The statistics of the duration of talkspurts and

silences in conversational speech are given by Brady [23],[24].

Voice parcels are assembled into packets by the speech terminal.

In waveform coders, the speech is not partitioned for analysis.

Rather, each speech sample is represented by a fixed number of

bits (for example, PCM). The speech terminal forms a packet from

a fixed number of speech samples.

Packets formed from voice parcels or voice samples are

transmitted through the network. If more such samples or parcels

are included in one packet, the delay in pa eke t .i za t i on (i.e.

waiting time for a packet of encoded speech) is increased. Other

delay parameters such as network delay in transmission and the

packet processing delay at the receiving station depend mainly on

the network topology and communication protocols used. As pointed

out by [11], the voice quality is greatly affected by the gaps

between packets (that is, by the variance of the total delay at

the output) and the average voice packet delay. Chin [11] triec

to decrease the coefficient of variation of the voice packet

delay by proposing an adaptive multiple access protocol while Dor

[9] tried to minimize the average packet delay.

In order to minimize the total delay, we have to decrease

the voice packet size so as to minimize the delay in

packetization. However, decreasing packet size in a CSMACD

network means an increasing of the normalized propagation delay

a1 and resulting in the throughput being reduced. With throughput

reduced, the network queueing delay will increase drastically.

Consider a cable of data rate 10 Mbps and 10 Km in length.

The propagation delay is about 50 us. A parcel of 20ms speech

coded by 3 2 Kbps Al) PCM [27] contains 640 bits of voice data.

Adding 40 bits of header, a nacket is formed con tain inn 68 0 bits.

The normalized propagation delay a' in this caase is 0.73, which

gives a maximum throughput of 0.2 using any CSMACD protocols.

The maximum number of calls that can be supported on the network

is 0.210 Mbps32 Kbps or 62.5. Moreover, the delay performance

is unacceptable as the network is easily overloaded.

We may form a packet from a number of voice parcels and adc

a header to it. In this case, the normalized propagation dela

would be smaller, the throughput is increased but then the dela

in packetization is also increased which may be unacceptable. For

example it each packet contains 10 voice parcels, a' will be

equal 0.078 which corresponds to a maximum throughput of OA.

using any CSMACD protocols| L 5]. The packetization delay however

is in c re ase d t o 0.2s.

Maxemchuk [16] proposed a variation of the CSMACD protocol

that yields movable TDM slots in an integrated voicedata local

network. With this protocol, data sources use conventional

CSMACD to access the channel, while the voice sources use a

variation of the same protocol. By requiring periodic sources

(voice) to generate periodic packets (each packet contains only

one parcel of 20 ms speech) after they have successfully acquired

the channel, the periodic packet transmissions are separated by

fixed time intervals. This results in a decrease in the network

delay. Channel contention is further reduced by requiring

periodic sources to transmit all of the data they have acquired

whenever they transmit and to schedule their next transmission a

fixed time T after their last successful channel acquisition.

The above protocol however is not suitable for networks with

a long channel and operating at high bit rate since more overhead

bits are necessary to be transmitted per voice or data oacke t

which are themselves quite short. To illustrate, for a cable

length of 10 Km and data rate of 10 M bp s, the preempt overhead

bits required for a 640 bits voice packet is 1000 bits [16].

Moreover, the size of data packets must be smaller than voice

packet which is quite small, say 640 bits (20ms parcel). Lone

messages must be cut up, with header added, and sent packet b

packet. If only voice traffic is present in a 10 Mbps, 10 Kn

network, the rn a x i m u m t h r o u g h put is only 640(1000+ 640) or 0.3

and the maximum number of calls that can be supported is 12.

(0.3 910Mbps32Kbps).

4.2 Voice Traffic in CLAR-MC

The network topology proposed in this thesis is particularly

suitable for voice traffic and telephone conferencing for its

nearly constant delay performance over a wide range of

throughput. The network access protocol (CLAR-MC) for the global

channels can be easily modified to implement voice-traffic and

telephone conferencing. A1so 1ow speed virtua 1 c i rcuits may be

implemented which are particularly useful for low speed data

communications. Figure 4.1 shows a schematic of circuitry

required to implement voice traffic and telephone conferencing

onto the M-channel LAN.

The modified CLAR-MC network access protocol is described

below:

4.2.1. Channel Status Table

Each station keeps a table recording the current information

on g1oba1 c h anne1s. A typica 1 tab1e looks a s follows:

C hi a n n e 1

N u m b e r

BusyReady,

IdleVoice

Tx Station

Number of

Circuits used

R x St:at: ion

Max number of

Circuits Allowe d

(

B

TJ

R

o

•

V

V

1

i;

id

rC

«

25

2 5

Fipiirp 4.1• Voice and Conference traffic transmission on M-LAIN

The Channe1 Status column tells whether the channel is

idle, busy, ready or being reserved for voice traffic.

When a talk spurt is generated, it is partitioned into 20 rn s voice

packets and sent out one-by-one through a reserved channel using

Maxemchuk's protocol. The maximum number of talkspurts that can

be reserved on each data channel depends on the channel

bandwidth, the normalized propagation delay and the voice data

rate. For a 1 Mbps cable and 32Kbps voice ADPCM encoding data

rate, we can a 1 1ow approximately 25 talkspurts transmitting

simultaneously on a global channel with nearly constant

packs tization delay of 20 m s and network transmission delay of

80 us. Define voice-circuit as a virtual circuit on a channel

that can support a talkspurt transmission, or a 32Kbps virtual

circuit in the above example.

If a channel is assigned for voice traffic application, the

third and fourth column of the channel status table will show the

total number of talkspurts circuits assigned and the maximum

number of talkspurts circuits available respectively. However if

a channel is transmitting data, the third and fourth column will

show the transmitting station number and the receiving station

number respect i. vely. For the table shown above, channel 1 is

busy and is used by station 15 and station 6 for data

tranmiss ion. Channel 2 is idle and channel 3 is ready.

Channel 8 and channel 9 are used for voice traffic application

and the maximum number of talkspurts circuit allowed on each

channel is 25. Thirteen talkspurts are n o w transmitting on

channel 8 while eighteen are on channel 9.

4.2.2. Voice Call Setup Procedure

There are three s t e p s involved i n t. he calling proce d ure.

First, the voice call is set up by a call-setup packet and an

acknow legement packet. Second, stations communicate with each

other by transmitting talk spurts through a voice channel

through reservation. Third, the conversation is terminated by a

call-release packet. The format of small signalling packets for

voice traffic (e.g. call-setup, call-release, talk spurt

reservation, talkspurt release, start-conversation etc.) is shown

in figure 2.7c. We describe in detail the voice call algorithm

below:

1) The calling station sends a call-setup packet to the

receiving station through the signalling channel. The

receiver station, if prepared, will give a ring to notify

the user. Wh e n t he user picks up the phone, a CALL-ACK

packet is sent back to the calling station to indicate that

the conversation may start. When a talkspurt is uttered by a

user, it is transformed into a periodic stream of voice

parcels in the station. At the same time, the station

searches for the highest number voice channel that has a

free voice-circuit, If none exists, it will select an

idle channel. A voice-circuit reservation packet is then

sent through the signa 11 i ng channel to notify the other

users that it has reserved the selected channel for

transmitting the voice packet stream. The corresponding

Channel Status Table's entry is updated by all stations.

Palkspurtis a ro encoded into fixed length voice packets

depending 〇»i tlie voice terminal encoding rate. Each packet

contains only 20-25 ms speech. Except t:he ” f i r s t，' voice

} ) a < ，k e t ： 7 G x l: ra bit：s c.re ,-ulclorl l)e[〇 re the packet:h e a d e r on all ”p e r i〇dic” v〇ice pdckef：s ‘ The d c c e s s p r〇t〇c〇1

is deigned so Lhat: thire is no collision between "periodic”packets. If a "first" packet [：inds itself collided with a

”p e r i o d i c” packet, it stops transmission immediately and

〇nly t：h〇 deader of i：he "periodic" packet ruined.

The Mp e r i 〇r]ic,' packet can condinue its transmission without

any loss of data. Figure 2.8 shows tlie formats of; M tMr s t M

voice packet and sul)seqaent "periodic" voice packets.

After the re se r t： i 〇a of a ta l kapu r t-c i rcu i t in a channel,

the ca l line] i on f)err〇 rms [；}]〇 ["ol lov/inq：

a) Jt ：n〇 ( t o r：s the assigned data channol until it is idle.

b) 1 L Uien the voice parcel and detects

for collison. If collided, it immediately stops

transmission and reschedules the transmission in a

random later time u^ing conventional binary exponential

backo£ L： sche,ne f-17 1 and repeat the trial procedure

u n I: i 1 s u c c e s s .

2 r»

It then marks the time of success and calculates the

next time interval T to transmit the subsequent

periodic voice packets. Since the packets are

generated periodically, short term TDM slots are in

effect being assigned for transmitting packets of that

talkspurt, and there will be no collision with

periodic packets of other talkspurts [16] (See figure

4.2a).

d) A station may find that the channel is busy in

transmitting other packets after the elapse of time

interval T. It has to wait until the channel is idle.

(See figure 4.2b). New coming voice samples during the

waiting time is included into the voice packet and

transmitted as soon as the channel is idle. If a

periodic voice packet is collided with a first

packet from a new talkspurt, only the preempt header is

ruined a nd it can continue the transmission

successfully neglecting the detected collision. (See

e)Voice packets are collected in a buffer at the

receiving station. They are then disassembled in tc

voice samples and passed to output register. The voice

samples are then decoded back into analog signal anc

output to the speaker.

A1 B1 A 2 Data Channe1 (D.C.)

[bl S igna 11 i rig Cha nne 1 (S. C.)

a) After a T sec. interval, station A succeeded in transmitting

the 2nd packet A2.

A15 CI A16

D.C.

c

S.C.

b) Station A has to wait for a short time t (tpacket tx time)

before the data channel becomes idle.

Collided Preempt

Header

A3 X A4 D1

D.C.

dc rO• KJ.

c) Station A transmits the packet but collided with a

first packet Dl. It insists on transmission and has its

preempt header ruined. The first packet withdraws itstransmission when collision is detected.

Figure 4.2: Transmission of talkspurt packets on the 'Voice. channel

a rb Ic d reservation packets for talkspurts

f) At the end 〇(； a talkspurt, the speaking station sends a

circii L l：- r：0 Le.：w3e [)ac 1 e t through the ：5 1 9 na 1 1 i ng channel

to notify all stations that it is releasing the

<ivS〇 " vo ice-c i r:c u i t " . All s t;a t i on s up(]a t p t: he i r

Channel Statues Tables ic.f；ording.ly. After receding the

cirxuit release pricket， the call receiving s ta t ion

sends back a Mcircuit-releaseM ACK to the calling station.

3 ) At end of a conversation, the calling station sends a

cal 1-relea^o packed to the destination st，U:ion„ The

destination sLal.ion sends back a RACK packet to the calling

s L a I: i on • T h g c a 11 i s te rm i n a t ed •

figure 4.3 shows a flowchart summarizing the procedures for

〇<-LLinfj i][) a v<;i<;e anci Cigure 4.4 sumi(MrL/,e.s tlie ,ste[)s C〇 r

an active user to Lransmit: voice packers of a talkspuirt.

The receiver ring theuser nni return an ACK

whe n the user a cc e pts(he call

Wai. t

Ca 11ingProcedure

» I,;• f!Voi c cChannel

Avai]ahle?

Transmit a

Call SetupPacket to

des t: ina t ion

Wait Tor RACK

paeket fromdes t ina t.ion

Transmit andReceive Voice

Ta1kspur ts

Transmi.t; aCa 1 1- Re 1easo

Pa eke t:

(Ca11 Terminat

Figure 4.3: Flowchart summarizing the procedures

for a station to set up a voice call

TalkspurtTransmittingProcedure

Search for

ilighest NumberedVoice channel

w i t h i d 1 eta 1 k firm r t e i rr 1i t

Search for anIdle channel to

assigned as voicec 11a n n e 1

Transmit a ta 1kspurtroservati.on packetthrough signal ling

channe1

Wai t f o rRACK

Wait untii thes elected cha n n e1

i s i d 1 e

Withdrawtransmission

s e t t r i a 1 p r o Iusing binarybackoff alpor

Col 1 i s i or Transmit the

f i r s t vo i.ce packe t

Suecess

Delay T i.ntervafor next packet

t: ransmiss i on

Wait untilthe channel

idle

Transmit the

voice packetdisregard an

co11ision

hast packett ransmi11 ed

Send a Talkspur IKelease paeket

thro' signal l injtchanne1

tie rmi na t

Ki'Mi re 4. A: Flowcluvrl for t ninsnri. It Lng I alkHpurt packolis

•i•-- 3. Telephone Conferencing Setup Procedure

Due to the broadcasting nature of the data channels the

network can also support telephone conferencing. If a station

w a n t s to setup a telephone conference with a set of other

stations, we call them the destination stations, it sends a

conference setup packet (see figure 2.7d) to all stations on

the signalling channel. The conference setup packet contains a

conference code for identification and„ the addresses of all

destinations. A subset of the destination stations is specified.

They are the major conference participants and are required to

send back ACKs to the conference initiating station. After all

conference ACK packets are received, the initiating station

sends a start-conversation packet to all stations to notify

that the conference call may start. Talkspurts generated by

speakers are processed and transmitted in the same manner as

outlined in the previous section except that the conference code

is included in all voice-circuit reservation packets and voice

packe t.

Only talkspurts from any two stations within a conference

may be transmitted simultaneously in the data channels. At the

receiving end, each conference station needs two analog vocoders

(figure4.1) that can receive talkspurts from any two stations

simultaneously. These talkspurts are mixed analogly at the

output speakers so that all participants can hear an interrupted

speech when two or more speakers are talking at the same time.

One of the dominating speaker will continue while the others

will yield and stop. E a c h pa rticipa tin g station main tains s

record of the number of active speakers and refrain from

transmitting talkspurts when two talkspurts are already on the

channel. Figure 4.5 summarizes the procedures of setting up a

telephone conference.

Conference

SetupProocdure

Voice ChannelAvailable?

Assign aConference Code

Transmit aConference Sotu

Paeket to a 1 1Iestinations

Wait for alequired RACK

Transmit a StaConve rsation

kel

Transmit amReceive1 voice

spurls

Rece p t i o n T r a nsmi s s i.(of a successful

Con f erence- torm i.na tckcl

and Procedure

Figure 4.5: Flowchart for Telephone Conferenc

Setup Procedure

4 * 2 * * 4 * ^〇w - 8 p e e d V i r t u a l C i r c u i t R e s e r v a t i o n

1 f d st:at,-〇n w a n t s {：〇 use a low speed virtu al c i r c u i t for

c-ontmuoiKs da ta t r a n s m i s s i o n , it can reserve a nv o i c e M c i r c u i t on

< h .、 丄 化 1 (十 < ⑴ ⑴ 、1 l() (!()〜）. " ： firsi hv; r〇 t、<K十 ⑴

s m a l l e r size p a c k e t s (640 b i t s / p a c k e t ) and then t r a n s m i t s these

p a c k e t s p e r i o d i c a l l y as if they were d i g i t i z e d voice packets.

E i r o r d e t e c t i o n / c o r r e c t i o n c o d e s are need e d on each packet： so as

to m a i n t a i n the data integr ity. This s e r v i c e s h o u l d be well

S U l t G d t〇r ]〇w ~ - i ^ 〇d data c o m m u n i c a t i o n s ince lower cost

i n t e r f a c e and less b u f f e r m e m o r y is required. V i r t u a l c i r c u i t o of

data rate up to 32 K b i t / s can be set up in this manner.

4 ，3 Pre 1 1 m i na ry Pe rf：o ruicj nee K va 1 no f： i 〇n

T h e r e are ma ny a d v a n t a g e s that make the M u l t i c h a n n e l n e t w o r k

p a r t i c u l a r l y ^ui table 1; 〇 ,； voice traffic and con l:e renc ing ：

1 ) 〇Ue 1:0 reservaliion [Protocols r：he b r o a d c a s t i n q

N，lL'n "1/ ,u f- call and con f；e renc^ incj call can be

^ u f)p〇rf;e(i. s t ation can reserve ddl：a c h a n n e l s to

t r a n s m i t its own talksp urt . As u s u a l l y one s p e a k e r is

flC f： 1 Ve i-n any con ve rsa C i on , TAST a d v a n t a g e [2S] can

o e e x f) 1 〇 i f： e i n t h i. s r 〇 t o c: 〇 ].,

2) The -ielay is very small and n e a r l y constant .

™ i 3 is D W t x c u U r a d v a n t a g e o u s f〇r voice

np() 1 i Cri I- ions .

30

Different types of service may be implemented easily.

For example we can set a limit to the total number of

channels to be used for telephone call we may allow a

station to reserve more than one voice circuit for

periodic data transmission such as fascimiie, video

etc. This is essentially virtual circuit switching.

When compared to the single cable CSMACD protocol for

transm i 11 'i n O VO i C e, we obse r ve I h e f o 1 1 ow i n a a d ra n h a n e a

lThe time required to detect collision in a network depends

on the end-to-end propagation delay and hence cable length

of the network. For the same collision detection interval

the number of corresponding preempt bits for a multichannel

network is much smaller compared with the single cable

network since the data rate is reduced by M times. The

throughout is thus i in d roved.

Maxemchuk [16] concluded that the throughput is increased

for an increased percentage of periodic (voice) load. In our

case, data and voice are separated on different channels,

therefore the throughput of the voice channel is not

degraded by the data traffic.

There is no need to cut data messages into very smal

packets to suit the requirements of voice packets. The

normalized propagation delay is decreased and hence the

throughpu t is increased.

dAThe maximum number of talkspurts in a data channel is

controlled through reservation so that channel stability is

ensured-

As an example, we consider the following Multichannal

network using CLAR-MC protocol:

Number of Channels M- 10

Total channel rate~ 10 Mbps

Data rate per channel- 1 Mbps

ADPCM encoding rate= 32 Kbps

C a b 1 e 1 e n g t h- 10 K m

voice parce1 size= 20 m s (640 bit s

overhead bits= 164 bits (including 108 bits preempt header

40 bits address header

16 bit CRC checksum)

average talkspurt length= 1.23 s (39.4 Kbits, 61 parcels)

average silence length= 1.77s (not encoded)

Consider a 20 ms frame in the data channel, the total number

of bits that can be transmitted is 20 Kbits. The packet size

equals- to 640+164 or 804 bits. Thus the total number of packets

to be accomodate in a 20 ms frame is 25, assuming an ideal

scheduling of voice parcels in the data frame. Since one channel

is used for signalling and reservation in ChAR-MC protocol, there

are 9 data channels. The total number of calls that can hole

simultaneously on the Multichannel network is 225, which is much

larger than that of single cable using CSMACD with the same

total channel rate. The actual number of calls that can be

supported is oven larger since we can obtain the TASI advantage

in this protocol. Therefore we conclude that the Multichannel

computer network with the CLAR-MC protocol is well suited tor

voice traffic applications. To compare the M-CSMA protocol with

the CLAR-MC protocol for voice application, we want to point out

that M transceivers are required there but only two is required

for CLAR-MC. Also in M-CSMA some complicated protocols must be

devised in order to implement virtual circuit switching based

on packet communications, and control the maximum number of

ta 1 ksou r ts us i ng the sane da t-a channe 1.

CHAPTER 5

ANALYTICAL MODEL

In this chapter we are going to analyze the throughput-delay

performance of the M-Channel LAN using the CLAR-MC protocol. It

is difficult to devise an analytical model that can describe the

complete network operation including reservation contention,

release contention, deterministic lookahead interval,

de terminis tic messag e sL ze etc. Therefore, we bui1d here on1y a

simplified mode 1. which can (jive a lower bound on the performance

on the network. The model is a two dimensional discrete time

Markov chain which can be solved by the iterative and recursive

matrix methods for the throughput and the average message delay.

5.1 Assu inp t ions

The following assumptions are made in modeling the network.

i; There are N independent stations communicating over an

M-channel network. The stations are all synchronized on

the s i g nail i. n g channel and the signalling packets (e.g.

reservation, release etc.) are transmitted on it in

s rn a 11 t i m e s 1 o t s.

2 A st ation is REA D Y w h e n i t finis hes its pr e vio us

transmission and starts generating new messages at a

p-icki1!;;;! lot. The message generation interval is

t he re fo r e g e ometrie a11y dist ribu ted. A f t er it ha

generate a message, it waits until its message has bee

3) A station becomes QUEUED immediately when a new message

is generated. It then waits for an idle or ready

channel if none exists. If one exists, all QUEUED

stations will contend for that channel. The probability

of success in contention during each time slot is

assumed to be the same for all contenting stations and

equals to u. For a given number of stations and number

of channels, the contention success probability is

estimated for various traffic rates by simulation. The

value of u used in the following queueing analysis is

taken to be the smallest estimate of the contention

success probabilities. This would give a conservative

estimate of delay-throughput performance since the

actual contention success probability is larger.

4) After a successful contention, a QUEUED station becomes

a RESERVED station and enters a reserve queue to wait

for the ready channels to become idle. As soon as a

channel becomes idle, one of the RESERVED stations gets

the channel and starts transmission.

5) Transmission time in the channel for each message is

assumed to be geometrically distributed. That is, each

transmitting station completes its transmission with a

constant rate at each time slot. The average

message length therefore is

6)We assumed that the lookahead interval is sufficiently

long that release contention always completes before

the end of the interval.

7)Ihe transmission of the RACK packets is not modelled

since a time slot is always reserved for it after each

successful reservation contention.

5.2 The Network Model

The multichannel network can be modelled as the closed

queueing network shown below in figure 5.1:

Arrival

Sec tion

Contentior

Sec tion

Channel Section

Figure 5.1: Queueing Model for the M-Channel LAN

with CLAR-MC protocol

There is a total oI: N customers circulating in the network.

We divide tiie network model into three sections: the arrival

section, the contention section and the channel section.

1) The channel section contains M parallel servers representing

the M data channels and a queue of size M which can accomodate up

to M customers that have reserved a channel. Let be a random

variable representing the total number of customers in the

channel section. In other words, is the number of stations that

are either using a channel for transmission or having reserved a

c 11anne1 for their next use. It is obvious that

2) The contention section consists ot a contention queue

containing QUEUED customers and a contention server. Whenever

there is room for more customers in the channel section (i.e.

the contention server starts working. In other words, it

Ci I 1 At JC f KQUEUED customers to content for

the channel faci1ities. The probability of contention success

under Phis r i r e 111 n ,s I a n r e s is u_ W h e r the contention server

stoos work inn.

3) The arrival section: There are i READY customers in the

arrival section. Each customer generates new messages at a rate

of packets per time slot

With that, the above model can be easily seen to be a two-

dimensional Markov chain with state vector (n,m) and state

dependent transition probabilities where n is the number of

QUEUED customers and m is the number of RESERVED plus BUSI

customers. Ai slate( n, in), there are three possible events whirl

can occur simultaneously at a certain time slot:

1) Generation of i new messages: the state transition in

this case is( n, in) ( n+ i, m)

2) Service completion of j messages: with state transition

( n, rn) (n,m- i)

3) Contention success: with state transition

( n m) ( n -1 m +1)

Let r= Prob[i messages arrival to the contention sectioni

in a slotJ. Let the time slot is small enough so that the

probability of four or more arrivals in a slot is negligible.

This is justified by actual data calculated during analysis. We

therefore have

Let qj be the probability that j messages have finished

transmission in a slot. Again, with time slots small enough, the

probability of four or more departures from the channel section

in a time slot is assumed negligible. We have

Let S(n,m) be the probability of contention success during a

certain time slot when the system is at state(n,m), then

S( n, m] for m 2M and nC

o the rwise

Then the transition probability p from one state to another

state is cii ven by

p[( n, m)-( n h-i- .1, rn- j +1) r| q• S( n, m) i, j: 0 ,1, 2,:

for a successful contention at the time slot.

For an unsuccessful contention, we hav(

D r(n,m)- (n+ i,m-i) : r• q•[ 1- S( n, m)] i, j: 0 ,1, 2,

5.3 Calculations

After we have built the Markov Chain, we can use matrix

resursive methods such as Gauss-Seidel Iteration or Successive

Overrelaxation[26] to calculate the equilibrium state

probabi1ities (n,m) of the Markov Chain. The throughput S and

averaqe delay T tor a aiven set of narameters can be calculated

from as foil ows.

Let a( k Prob [k channels are used], M, and

b( j) Prob [the reare jmessages in contention sectionand

channel section]

Then,

From (1), we can obtain the average number of channels in

nQOfl A ci

Define Ns as the average number of stations that are either

in the Contention Section or Channel Section and the average

delay T as the average length of the time interval between the

instance when a message is generated and the instance when it

finishes transmission. Since we are considering a closed queueing

network, the rate of message generation at the arrival

section is equal to the throughput at the Channel Section, or

where A is the rate of the channel server. From

(2), we have

The average delay T by Little's formula is

For an M data channel and one signalling channel system, the

overall system utilization S is

5.4 Numerical Results

We assume a total channel capacity of 10 Mbps which are

divided into M+l channels of 10(M+1) Mbps each. By varying the

message generastion ratewe can calculate the throughput-

delay characteristics for each value of M.

We plot the re su Its for cases of M--3, 5, L 0 ,1 5 in figure 5.2

and compa red t o the s imu]a tion resu1ts. As ex pected we found that

the results are very close to each other and the performance

from the analytical results is a lower bound of the throughput

performance and an upper bound for the delay performance.

Particularly we observe that the characteristics of analytical

results and simulation results merge together at high traffic.

This is because the probability of success contention if

approximately the same. At low traffic the analytical

characteristics slightly deviate f rom the simulation results a f

the probability of success contention is underestimated.

Figure 5.2: Delay-Throughput Characteristics of OLAR-MC protocol

Comparison of analytical and simulation results.

Message size is geometrically distributed.

CHAPTER 6

SIMULATION RESULTS

We use GPSS-V to simulate the various operating conditions

of the proposed network topology. As expected, we found that a

steady maximum throughput is maintciined under very heavy loading

condition. This maximum throughput moreover is independent of the

number of stations connected to the network. We summarize the

results and observations below:

Figures 6.1 and 6.2 show the delay-throughput performance of

the CLAR-MC protocol on a multichannel network with the following

ooeratina parameters:

Total no. of channels

Data rate per channel

Total Channel Bit Rate

Cable length

Propagation Delay a'

Reservation packet size

Slot size (2a+32)

Message size

Lookahead Interval (LI)

10

1 Mbps

10 M bp s

10 Km

50 us

32 bits

13 2 us

50 slots (825 bytes)

20 slots (165 bytes)

Message transmission time in a single 10 Mbps cable is

Normalized propagation delay is

Also shown in figures 6.1 and 6.2 is the delay-throughput

performance of a single CSMA channel of capacity 10 Mbps and

multichannel M-CSMA protocols with the same operating parameters

for comparison. We found that the CLAR-MC protocol has a higher

maximum throughput 0.9 compared to the single channel CSMA

Figure 6.1Throughput comparisons between CLAR-MC and other protocols

Throughput S versus Offered Traffic G.

Figure 6.2 : Delay-Throughput comparisons between CLAR-MC and other

protocols.

protocol 0.6. The delay performance of the CLAR protocol is

similar to the M-CSMA protocols at low utilization. But when the

offered traffic increases beyond the network capacity, stable

maximum throughput is obtained for the CLAR-MC protocol. The

CLAR-MC protocol therefore is more suitable for larger networks

under heavy loading condition. An added advantage is that only

two sets of transceiver are needed in the CLAR-MC protocol, while

M transceivers are required for the M-CSMA protocol.

Figure 6.3 shows two sets of throughput-delay

characteristics of CLAR-MC as a function of M and a (normalized

propagation delay) for both fixed and geometric message sizes.

The maximum throughput obtained for geometric message size is

slightly smaller than that for fixed message size. This is

because for geometric messages, some messages will have sizes

smaller than LI (the lookahead interval size). The transmission

may terminate before any successful reservation can be done

through contention and results in the data channel being idle for

a certain period.

Figure 6.4 and 6.6 show the effect of LI on the delay-

throughput characteristics of CLAR-MC for fixed and geometric

message size respectively. From figure 6.4, it can be seen that

for fixed size messages, the delay-throughput performance is

nearly independent of Li, and any value of LI greater than 5

slots can be chosen to obtain the maximum throughput. Figure 6.5

shows that for geometric messages, the throughput is more

dependent of LI and a larger LI is needed for maximum throughput.

Figure 6.3: Delay-Throughput Characteristics of CLAR-MC protocol

as a function of number of channels M and average

message size B. Comparision of fixed and geometrically

distributed message size.

Figure 6.4Delay-Throughput Characteristics of CLAR-MC protocol

as a function of lookahead interval (LI).

Message size is fixed.

Figure 6.5

Delay-Throughput Characteristics of CLAR-MC protocol

as a function of lookahead interval (LI).

Message size is geometrically distributed.

Figures 6.6 and 6.7 show the delay-throughput

characteristics of CLAR-MC as a function of message size. We can

observe from figure 6.6 that t he data channels can be fully

utilized for fixed message sizes greater or equal to 50 slots,

which is very close to the B=(M-l)nP result derived in chapter

3. Again, we see in figure 6.7 that for geometric messages, the

throughput is more sensitive to B compared with fixed size

m pssanp.q.

From the simulation results, we conclude that the CLAR-MC

protocol has the followina orooerties:

1) Stable maximum throughput can be obtained.

2) Constant delay performance can be obtained over a large

range of through) p u t:.

3) The throughput-delay performance for fixed size messages is

better than that for geometric messages.

4) There is a lower bound on message size in order for fully

utilizing the data channels (B- (M-1)n P). See chapter 3

for a more detail discussion.

5) Maximum throughput is attained for any LI larger than 10

slots. LI surely cannot be larger than the message size.

Figure 6.6 Delay-Throughput Characteristics of CLAR-MC protocol

as a function of fixed message size.

Figure 6.7: Delay-Throughput Characteristics of CLAR-MC protocol as a

function of geometrically distributed message size.

CHAPTER 7

CONCLUS IONS

In this thesis we have proposed a new local area network

topology and a new communication protocol to be used on it.

Analysis and simulation results show that the multichannel

computer network with local and global transceiving media is a

feasible design with many nice features. Although there is a

constraint on the message size, we found that even in the case of

10 Mbps, 10 Km, the average message size needs only be 825 bytes,

for maximum throughput in a 10 data channel LAN. Fast circuit

switching can be implemented on the network so that a station,

once acquired a channel, can transmit the entire message to the

destination station. This implies a longer average message length

per message transmission when compared to the single CSMA case

where messages must first be broken up into packets.

We have also shown that the proposed multichannel network

and protocol CLAR-MC can offer an integrated service to different

applications. It can support message switching, packet switching

as well as fast circuit switching. It can provide services such

as message data transmission, two-party telephone calls,

multiparty telephone conferencing, fascimile as well as providing

virtual circuits with customized data rate.

REFERENCES

[1] Wi 11 i am S t a 11 i ng s, L oca 1 Networ k Pe t:f ormance, I E E E

Cominun i.a a t ion 8 Magazine, Fgb. 1 9S4, Vo 1. 2 2, No. 2, pp. 27-36

r 21 Robert: Bowerman, Choosing a Local Area Network Interface

Age, July 1983 pp.55-59, pp.138.

T 3 1 M. Ajmone Marsan and 1). R of 1 i n e 1 la, Multichannel Local

Area Network Protocols IEEE Journal on selected areas of

communications, Nov.1983, pp.885-897

[4] M. Ajmone Marsan and 0. Roffinella, Nonpe r s .i s ten t M-CSMA

Protocols for Multichannel Local Area Networks, CH 18100-

23200000100 IEEE 1982.

[5] Fouad A. Tobagi, Multiaccess Protocols in Packet

Communication Systems, IEEE Trans, on Comm., vol. corn-28,

n o.4, April 1980.

[e; Edmund A. Earrington, VoiceData Integration using Circuit

Switched Networks, IEEE Trans. on Comm., vol. com-28,

no.6, June 1980.

[7: Myron J. Ross and Osama A. Mowafi, Performance Analysis of

Hybrid Switching Concepts for Integrated VoiceData

Communications, IEEE Trans. on Comm., vol. corn-30, No. 5,

M a y 1 9 8 2.

TBI John G. Gruber, Delay Related Issues in Integrated Voice

and Data Networks, IEEE Trans, on Comm., vol. com-29,

No.6,Ju n e 1981.

[9] Don H. Johnson and Gerald C. 01 1e a r y, A local Access

Network for Packetized Digital Voice Communication, IEEE

Trans, on Comm., vol. corn-29, No. 5, May 1981.

n nlGilbert F. Williams and Alberto Leon-Garcia, Performance

Analysis of Integrated Voice and Data Hybrid-Switched

Links, IEEE Trans, on Comm., vol. com-32, no.6, June 1984.

n iChin T. Wu and Vanan R. Dhadesugoor, An Adaptive Multiple

access protocol for Integrated VoiceData Local Area

Networks, 83CH1956-2830000-0959 1983 IEEEE.

C. Petitpierre, Influence of the Network Topology upon th

Throughput Limitations, CH 2028-8840000-0428 1984 IEEE.

T13 Michael J. Ferguson and Lome G. Mason, Network Design for

a Large Class of Teleconferencing Systems, IEEE Trans, on

Comm., vol. com-3 2, no.7, July 1984.

r l a| Jon W. Mark, Distributed Scheduling Conflict-Free Multiple

Access for Local Area Communication Networks, IEEE Trans,

on Comm., vol. corn-28, no. 12, December 1980.

[15 Leonard Klein rock and Fouad A. Tobagi, Packet Switching in

Radio Channels: Part I-Carrier Sense Multiple-Access Modes

and their Throughput-Delay Characteristics, IEEE Trans, on

Comm., vol. com-23, no.12, December 1975.

[16 1 N.F. Maxemchuk, A Variation on CSMACD that yields Movable

TDM Slots in Integrated VoiceData Local Networks The

Bell Syst. Tech. Jour., September 1982.

ri7iAndrew S. Tanenbaum, Computer N e t w o r_k, Prentice Flail

Software Series, 1981.

ns L. Klein rock, Queueing Systems, Vol. I, Theory; Vol. II,

Computer Applications, New York: Wiley Interscience, 1975.

[19] Charles H. Sauer and K. Mani Chandy, Computer System;

Performance D. 2' Prentice-Hall, Inc., Englewooi

Cliffs, N.J. 07632 1981.

r ?n Donald W. Da vies and Derek L. A. Barber, Communication

Networks for Computers, Wiley Series in Computing 1975.

r? iHisashi Kobayashi, Modeling and Analysis: An introduction

to System Performance Evaluation Methodology, The Systems

Programming Series, IBM Editorial Board 1978.

[22 P.C. Wong and T.S. Yum, A Multichannel Computer NetworF

with Loca 1 and Global Transceiving Media, IEEE INEOCOb

March 1985.

P.T. Brady, A technique for investigating on~off patterns

in speech, Bell Syst. Tech. J., vol.44, pp.1-22,1963.

[ 24 ] P.T. Brady, Effects of transmission delay o

conversational behavior on echo-free telephone circuits

Bell Syst. Tech. J., vol.50, pp.115-134, 1971.

[251 T a t s u y a Suda, Yechiam Yernini, Hideo Miyahara, Toshiharu

Hasegawa, Performance Evaluation of A Packetized Voice

System- Simulation Study, 83H1874-7830000-0749 1983

IEEE.

W.R. Daumer, X.Maitre, P. Mermelstein and I. Tokizawa,

Overview of the ADPCM Coding Algorithm, IEEE GLOBECOMf84,

SESSION 23.

r 27 I N. Psiinenatos, J. Gruger, G. Goddard and D. Mondor,

Network Considerations for the introduction of 32 Kbps

A D P C M Systems in Telecommunications Networks, IEEE

GLOBECOM'84, SESSION 23.

APPENDIX: Simulation Program in GPSS

hlhckMUMRFR LOC CP E nATIG N A, 0,C,D ,E ,F,G, H, I

CLRS-MC SIMULATIONCOMMENTS

NUMBER CF CHANNELS= 10JLI= 50MESSACE SIZE= 500» EXPONENTIAL, WITH FAST MESSAGE ALLOWEDNUMBER CF STATIONS= 25; INITIALIZATION OF MESSAGESFUNCTIONS AND VARIABLES

ST MULATEFU N C T I C N R N1 ,C2r

0,01, 12 FUNCTION R NO, C 2 4

0.0,00.1jO. 1 049 .2,0 .2220.3,0.3 550.4, 0.5 09.5, .690.6, 0.9150 .7,1 .20. 75,1.380.8,1.60.84 ,1 .830•88,2. 120.9,2.30.9 2,2.520. 94, 2.- 810. 95. 2. 990. 96, 3. 20. 97, 3. 50.98,3. 9 0.99, 4. 60 .9 95, 5.30.998,6.20.999,7 .00 .9R97, 8. C

1

4rz

67

VAR I A OLEVARIABLEVAR I ABLEVARIABLEVAR I ABLEVA RI A OLEVAR I AELE

2 50 0X 1 2X 2 250R1 05 00

DOUBLE THE EST. STATIONHALF THE EST.STATION

LOOKAHEAD INTERVALNUMBER OF DATA CHANNELS

CHANNEL NUMBER OF REQ.CHANNELMESSAGE SIZE

1E-j

4567,q

~cT

1 01 11 21 31 41 51 61 71 81 92 08 1o o_ z_.

2 42 57 67 72 87 93 03 13 23 3

TR ANSFFRC-E NFP ATETP ANSFFR

TR ANSFFRGENERATETR ANSFFR

V 1, FN2, 5-,LAB30V1 ,FN2, 5,LAB30V1,FN2, 5,LAB30V 1, F N 2, 5,LAB30

GENERATETP ANSFFRGE NFR ATETP. ANSFFRGENERATETR ANSFFRGENERATETR ANSFERGENERATETR ANSFERGNER ATETP ANSFERgenerateTR ANSFFRGE NFRATETR ANSFER

V1,F N2, 15,L A B 30V 1, F N 2, 15,LAB30V 1 ,FN 2, 15, LAE30V 1, F N 2, 1 5,LAB30V1,FN2,25,LA 3 30VI,FN2,25, L A B 3 0V1,FN2,25,LAB30V 1, F N 2, 2 5,LA 83 0

GENERATETR ANSFERGENFR ATP

_TR ANSFFR_generateTP ANSFFRGENERATETP A N SF ERGENERATE

V 1, FN2,35, L A B 2 0V1,FN2,35,LAB30V 1 ,FN2,35,LAB30V 1, F N 2, 3 5,LAB30V 1, F N 2, 4 5

STA TLMuf.i J (J Mtl L

1234567a9

101 i121 J1 41 5lo1 7ia1 92 U2 12223242526272a29303 132333 4353d3 7A H39404 1424344454 6474a4 9505152535455

3 43536373 33 n4 0

_ 4 14 24344454 6474 8495 05 152

_5 35 45 556

.5 75 a

5 9

6 06 16 2636 4656 6

_6 76 86 97 07 17 27 37 4.7 5767 77 87 n5 08 I3 2

.3 38 48 5

TP ANSFERGE NEP ATETP ANSFER

_G5N ER AXETR AN5FFRgenerateTR ANSFER

_GE NER ATFTR AN5FERGENERATETDAN5FERGTNFR ATFTP ANSFERGENERATE

.TR ANSFERGFNFR ATE_TP ANSFERASS TGNTP ST LFAS SIGN

LAB30

L AO 1 4LAB2 4

_.L A B 1_LAB4

MARKTEST NETEST NF

TTNKLS

LAB34 LI NK

L A B 1 1 AS SIGNTE ST FGATE LSAS S IGN...SE L EC TL!AS SIGNAS SIGNTEST 1BUFFERGATE NUTP ST NFTEST NE_

LABI 3LABI 6

LAD?

LABI 7SELECTLSCATF FVTABULATE.TEST....CCUEUCSF I ZrDEPARTI (DG ICRTEST NETEST NEADVA N fFRE LEA SE_TABULATETR ANSFFR

LA 03 1363 7

ADVANCERE LFASE

,LAG 30V1,FN2,45,LAB30

_V_ A_,EN 2-j_4.5_,LA030V 1• F N 2• 4 5,LAE30V1iFN2,55,LAB30V1,EN2,55,LAE30V1.F N 2,55, LAB30V1 ,F N 2,55, LAG30V T, F N 2, J£ 5,LAG309, V7 ,2P9i5 0L AB14

.JLD_,_I5

2,P7,LAB34P 10 i 1 »L4B42, 1, V5, ,LA.E6WAITljFIFC

WAIT2,FIPC

1 ,X1

2, P 7, L A B 1 62.2, LAB 1 31, 12,1,V5,|,LAE16I ,24,P 1 ,1P4, 1 T LABI 2

V 62• P 7, L A B 1 71 ,P10,LAB 172,1,V5,,,LAE6V 6,L AB74w1 A R R. n.1 APIrV 6V 6V 6p 1

1 iPIOILAB 232,° 7, L A B3 120

_V_6_1, L A B 5

10V 6

MESSAGE SIZE,EXPONENTIALSMALL MESSAGE?

_S FT MARKFQR FASJ_P_ACKETNEW MESSAGE GENERATE TIMEREL PACKETSMALL PACKET

A5SIGN_A READY CHANNEL EXISTSWAIT FOR CONTENTION INTERVAL

WAIT FOR CONTENTION INTERVAL

ASSIGN EST. NO. OF STATIONSA REL PACKET?PREVICLS REQ CM. IS COLLIDED,JUMPNO _READY CHANNEL EXI ST ,PRE.NOT. COLL IS

IF NCNEXIST READY CHANNEL,JMPSET PR0G=l2

CHECK IF BRANCH. DELAY T. X.

NO NEED TO WAIT FOR REL PACKET SNO NEED TO WAIT FOR SMALL PACKETS

RECHECK IF EXISTENCE OF READYCH.CHECK IF COLL.FLAG SET

SEIZE THE REQ.CH•AN ARTIFICIAL QUEUEINDICATE THAT THE REQ. CHANNEI IS__USE.DIFFERENT TREATMENT FOR SMALL PACK

FIXED REG.PACKET

RELEASE PACKETS =10 ONLY.SET FREE THE REQ.CHANNEL

5 65 75 35 56 06 1626 J646566A 7

6 8657'J71727574757677787 98 J8 18213 3

84858 68 7888 990

.9 192939495969 798QQ

1 00I 011 021 051 041 051061071031 091 1 01 1 11 1 2

n q

q o

OO9 1° 29 39 49 5

9697OR9 9.1 001 011 021 031 041 051 06I 07I 0 8

_1_0 91 1 01 1 11 1 21 I 31 141 1 51 161 171 1 01 191 20

1 2 11 2 21 231 241 2 51 26

1 8 71 2 81 291301 3 1

1 321 3 3

1 3 41 351 36137

LAD23

LABI 0

LAR7

LABI 2

LAOS

LAB26

LAB27LAR3

LA02 2

LA02 0

LA 02 1

L A B 1 9

LAB1R

LA GoLAB6

TR ANSFER

AOVANCERELEASETERM I NATE~LOG ICRFUNAVA ILADVANCETRANSFERDELAY TRANSADVANCETR ANSFER

SELECTNUGATE LRTEST NFASSIGN-SELECT NU'TR ANSFER

LO GICRCU EUFTA EULATESE I ZEDE PART

ASS IGN

AS SIGNADVANCEAS SIGNSn LITA D v A N c rRELEASETABULATETERM I NATE

AS SIGNTR ANSFERAS S IG NTABULATEPARKTR ANSFER

LB GIGSTA BUL ATETA ELLATEUNLINKTERMINATE

SUFFERLI NK

GENERATEGATE LPLOG ICR

LAD is

P 9V 6

2 0V 69V LA R24M I 5 S 10 N9

, L A E 2 42,1, V5,, ,LA E92,1 ,V5,, ,LA E3R2,LABSP 2, V 5,L AB 272+, 12,P2, V5 t,, L A B27,LA 0 2 62,1 ,V5,, ,LAE9P 2P 23

P2_P 2

P 9, V 4, L AB 2 19-, V43, V 4pg8 »P 21 ,L AB 1 9P 3P 29

3 ,P9, LAE227 ,20

1 1,LAP 14

p q61 1W AT T .LABI, A l.L

CHK IE NQ CHANNELWAIT,FIF0

1 0,,,, 120,LAB322 22, 1, V5,, ,LA E50

GOT C EEL• RF. AD Y.

ADVANCE THE SMALL PACKET

KILL THE SMALL MESSAGECOLLISION INDICATIONMAKE THE REQ. CM. UNAVAILABLEARTIFICIAL DELAY INTRODUCEDRE T R AN MIG SI0 N

ARTIFICIAL DATA DELAYRETRANSMISSION TRY

SELECT A FREE CHANNELNO UNUSED CH. —JUMP

UNUSED CH. IS READY? YES,JUMPLAST UNUSED CH. JUMP

RE SET THE RE ADY SWITCHJOIN CUEUE IF NECESSARYACTLAL DELAY FOR CH. ACQUISITIONSEIZE SERVER

SUBTRACT 200 FROM P9

SERVICE RECEIVEDSTORE THE CHANNEL REL NO.GENERATE A REL PACKETLOOKAHEAD INTERVAL TRANSMITTEDRELEASE THE CHANNEL ACTUALLY

SUCC. PACKET INTERARRIVAL TIMESASSIGN NEXT NE W PA CKE T 3

STORE VALUE OF TX

SET A RELEASE PACKETINTER ARR I VAL T I ME OF REL PACKITSET MARKER FOR RELEASE PACKETCONTENT I ON

RELEASE THE CHANNEL TO READY

EACH REL CHKS. NO. WAITING

K I L L TIT E REL.' PAC KET

AS5IGNFDA RT I FICIAL DELAY AD DED

CONTENT IC N INTERVALNO COLLISION? JUMPSET COLLISION FLAG FOR PEL. P

READY CHANNEL DOE SN•T E X I ST, JUMP

1 1 JI 141 i 5I lb1 1 7L 1 81 191 201 211 221 2 J1241 251 2b1 271 281291301 3 11321331 341351 361371381 391 401411 421 431441451 4-o1 471481 491 501511 52I 531 541 551 5b1 571 581591 601611 621631 6416 51 661 671681 69

1 381 3«1 40

_JL4.11 421 431441 43

1 4 61 4 7

_1 43I 491 50

•1511 521 531 5 4

1_551 56

1 57.1-58

1 591 601 611 62

L AR3 7

SAVEVALUETP. ANSFERLOG ICS

_SILL E.C X L_S_GATE LSSA VEV ALUETEST L9 A V FVAI I J-

LAD5 LOG ICSLOGICSJ A B U L A T_F_TA BULATETABULATEFAVA ILUNLINK•UNL INK-'.TERM I N AT'r

SENEP A TFTERM I N ATE

SAV

GENERATE_S AVE VALUE

AS SIGNLOG ICSLOOPTF RM T NATF

123_467

_89101 1

TA BLETA BLETA BLETABLETABLETA BLE

_TA ELF!TABLETA ELETA El E

ST ARTRE SETST ART

.VA R IA BLECL EARST ART

ST ARTVA P T A Rl FCL E ARST ART

.RESETST ARTVARI ABLECLEARSTARTRESET

1 f V 2,LAB5022

2 ,._1__YFL»_» _J_LAJEJ1Q2 1• LAB501, V3X1, 1 L AB5 01. 1

2 02 1

781 0V 6

_ W AIJ2,LAB 11, ALLWAIT1iLABl 1 ALL

_1_Q_Q_Q_1

0»«» 1» 3-1 ,1—

1,22P 11, SAV

I A, 0,10,30IA,0,10.30

_MP5 ,0, 100,40W$LABB,0,1,40M P1 1,0,10,20XI, 1,1 ,50

_CJdSJtAXT_L.A_Qj.JLj_l.Q_Q_MPS,0,100,40CHS VIA I T2, C, 1,10rnViATT.n_i .inn

~ 5 ,NP

1 0_2_2J5_0_

5, N P

1 0pnnn

5, N P

1 750

5. N P._

rni i t c Tnki--nni im F m

RESET COLLISION FLAGR E A DY C H A NN El DO E SM.LT._E X I ST., J.U.M PUSED, SUCCESS? JUMPIDLE, HALF THE NLOWER BOUND OF NCCTT M— 1

MO COLL, SET SWITCHIDLE ,S E T SWITCHTA BLE THE ESTIMATF VALUETABLE THE REAL NO.TABLE THE REL. PACKET WAITINGMAKE THE REQ.CH AVAILABLESET FREE THE. REL.PACKETS

IN IT. OF SWITCHES. I N I T.. GF N._._

INIT. PI, TO 12

nFr. RY. 1. TF 7FRF1. HI.TI DHP

1 7 01 711 72173I 741751 761771 76179180L 8 11 821 8 J1 841 85I 861871 88

_. 1 891901911921931 941 9P1 96

_ .1 9 71981 9 92 0020 12022032042 0520620 72082 092 102 I i2122152 142 152 1 62 1 72 182 1922022122222 5224225

1

1

STARTVARIABLECLEARSTARTRESET

VARIAELECL E ARST ARTRESETST ARTcm n

1 0i c; n n

c;. Ki n

1 0i nnn

5, N P

1 n

22 72282292 JO2 J 12 J22 J 3234238236237238