Multi-stage ATM Switching Systems - Keio Universitybiblio.yamanaka.ics.keio.ac.jp/file/conf_ieeeatm_nakai...system ArchitectlEe: Based on Highly StatisticalOpticalWDM hterco-ection,M

DistributedTra氏c Control Method for Tbit/s

Multi-stage ATM Switching Systems

Kohei Nakai, Eiji Oki and Naoaki Yamanaka

NTT Network Service Svstems Laboratories

3-9-ll Midori-cho. Musashino-shi, Tokvo 180-8585 Japan

tel: +811422-.59-3608. fax:+81-422-59-4819

E-mail: nakail◎nttssl.nslab.ntt.co.JP

Abstract

This paper proposes a distributed tra･丘･c control method for large multi-stage ATII

switching systems･ In the proposed switching system? each port of the basic switches hasits own trafBc monitor, and each line unit (LU) periodically obtains COngeStion information

about available paths from OA-1工cells in order to route a newvirtual circuit independently･

The perfomance of the proposed system depends on the intervalbetween OA-V cellSl Weshow how an appropnate intervalcan be determined in order to majdmize the number of

user cells that each LU can send.

1 Introduction

Asynchronous transfer mode (AT氾) is believed to yield the best high-speed multi皿edia infras-

tructure. ATM netWorks can handle various services, slユCh as high-speed data co皿皿1-ications,

real一七imevideo conferences, andvide0-on-demand (VOD)･ With demand for sllCh services ex-

panding, ATM networks will reql血e switching systems that offer Tbit/s throughputl1, 2]･

We previously proposed a scalable three-stage ATM switching architecture that uses OPtical

wm/I (wavelength division multiplexing) grolユped hnks and dynamic bandwidth sharing【3, 4ト

The former redllCeS the nlユmber of cables req-並red and the latter prevents the statistical multi-

plexing gain from fining as the switching system expand5･That switching system reqlLred explicit path selection･ When a newvirtualcirc血t is to be

added from some inp-ユt hne lmit (LU) to some OlユtPlユt LU, the system must丘nd an appropriate

path through the switching networkwith sd五cient mused bandwidth on each of its hdiS tO

accommodate the new virtual ci三･cuit.

A conventionalsysten has a central control processor which manages the bandwidth onall

of Lnks in the switching netWOrk･ The processor b･andlesall virtllalcircuits in the system And

i七五nds a potentialpa･th having enolユgh lmⅦed capacity･ However, it is difRcult for a central

control processor to handle all thevirtualcircdts when a switching system becomes large･ A

large switching system whose thro-ユghplユt is Tbit/s class, requires a distributed tra毘c control

method.

We propose a new distriblユted tra氏c control method for large multi-stage switching systenS･

In this switching system, each port of the basic switches hぉits own trafEc monitor･ And each

LU periodically obtains congestion infomation about available paths from OAM ceus in order

to route a newvirtual circuit independently.

The performance of this system depends on the intervalbetween OAM cells･ If we set it too

long, the congestion information in LUs is delayed, blユt if we set it too short, the nlユmber of

OAM cells in the system increa5eS･ We show how an appropriate inte-albetween OAM cells

can be determined in order to ma:imize the nlユmber of user cells that eacも. LU can send.

109

･/･ ･.十

WDM group

link ∃芸三.i

Figule 1: Scalable 3-stage ATM switching architecturewit五WDM grouped links

The remainder of this paper is as fouows･ Section 2 brie丑y reviews a scalable 3-stage ATM

switch architecture vith opticalWDM grouped links･ Section 3 presents a distributed trafBc

control method and shows how aL appropriate interval between OAM ceus can be dete血ned_

Finally? section 4 su皿皿arizes the key points･

2　Scalable three-stage ATM switching architecture with WI)M

grouped links

Figure 1 shows a scalable three-stage AT氾 switch architect-ewith opticalWDM grouped

links【3, 4]･ The expa皿Sion factor iS M tines the basic network, so the totalthIOughput of this

system is MN times that of the basic switch- For example, a system capacity of 5.2 Tbit/s can

be achieved using 8 X 8 80-Gbit/s basic switches (N - 8,M - 8).

Each stage consists of N switch groups aLnd each switch group COnSists of M basic switches.

We call the basic switch a membe, toコユOre Clearly explain our new switch architecture. The

(n,m) basic switch is the nth member of the nth group.Ea･ch basic switch has N input ports and N output ports･ Each port multiplexes M links

corresponding to the M wavelengths that are multiple=(ed into one optical五ber. The set of M

wavelengths is called a WDMgrOuPI

Each output port of the basic switch is connected to its intended switch group in the next

stage･ Each wavelength belonglng tO a WDMgroup of an output port is connected to its intended

member in its intended switch group Of the next stage by wavelength switching･

The bandwidth of the switch port is shared among each wavelength ill the WDM grOuP･ The

totalbzLndwidth of wavelengths in the WDM group is united to the speed of the switch port (C

bit/S)･ The bandwidth of each wavelength changes dynanically･ The statistical皿ultipleIdng

ga血 is not reduced even if the scale of the switching system is increased. When tile basic network

expands by M timeS7 the ma=bmm bandwidth of each血k (C bit/S) is not divided into C/M

bit/S.

AnexanPle of cell routing is shown in Fig1 1･ Let us consider the routing of an ATMconnection from the ars上 stage (1, 1) to the third stage (1,M). h the五rst stage, cells are routed

1 1 0

!

TrafBc moTlitor

Speed of

switch port C

hputLUs OAMJ Lt; Lt7 簸�ﾖ6Vﾆﾆ也FUEf�ﾅF��奉ﾂ雹Yl���WB�

Ltj LU

U __一一一■-lll.-I

LU 辻�版R�

LU LU LU ��版R�@-~■: 儉U U

_-■一一一一■一一■■■■■ �/LU

1 st stage　　　　　　　　　　　　　2nd stage　　　　　　　　　　　　　3nd sta書e

Figure 2: Distributed tra丘c control method for 31Stage ATM switching systems using WDM

grolユPed li延s

the Nth port in the (1, 1) basic switch, and converted to wavelength A1 and rolユted to the second

stage (N, 1), using Wavelength switching･ h the second stage, cells are rollted to the五Tst port

in the (N, 1) basic switch, and converted to wavelength入M in an optical五ber for rolユ七ing to the

third stage (1, M).

3　Distributed tra氏c control method for large multi-stage switch-

ing systems

Figlユre 2 shows oW proposed trafBc control met五･〇d applied to a three-stage ATM switching

syste皿5心1ng WDM grolユPed lin上s. h this switching system, each port of the basic switches

has its own tra缶c monitor. And each LU periodically obtains congestion information abollt

avaムable paths from OAM cells in order to rolユte a neWvirtualcircl云t independently.

h each port of the basic switches, the residualbandwidth is estimated by counting arriving

cens in a certain period〔5, 63･ When the residualbandwidth in a port is below B, bit/S, the

congestion information bit of arriving OAM cells is set to l･

Each input LTJ sends OAM ceus forward to output LUs every To seconds. The output LUs

retllm OAM ce]ユs to the inplユt LTJs. Each inplユt LU rolユteS Virtualcircdts independently lユSlng

the congestion information in OAM cells･ The r0-1七ing lookllP ln each inplユt LU inserts a path

speci五cation into the cell headers of arriving cells･

正we set To too long, the congestion information in LUs is delayed･ LUs rolユte manyvirt-ユal

circl止ts to congested ports in the system. The totalbandwidth assigned in a port exceeds the

speed of the switch port (C bit/S). We call this overbooking-

Figlユre 3 shows the relationship between Br and the overbooking probabihty. We must Set

B, to awide bandwidth so aS not tO exceed the speci五ed overbooking probabihty･ Figure 4

shows B, for assuring that the overbooking probability < 10-9･ If we set To to 1 Ⅱ旭, then Br

must be set to 0_9 Gbit/S. 1

0n the other haLnd, if we set To too short, the number of OAM cells in the system increases･

Tも.e rate of the OAM cells sent by each LU (Bo bit/S) is inversely proportionalto To, as shovn

in Fig･ 5. If Bo increases, the number of user cells in the system decreases.

1The rate of connections is 100 Mbit/S.

1 1 1

10-i

0　　　　　　0.5　　　　　　1.0　　　　　　1.5

BT [Gbit/S]

Figllre 3‥ The overbooking probabihty when B, increases

0.01　　　　　　0.1　　　　　　1.0　　　　　　　10

To､【ms】

Figure 4: B, for assunng the overbooking probability < 10-9

1 1 2

0

.

1

J

一

.

3

T

一

1

.

6

.

7

.

8

.

9

0

0

0

0

0

0

(

U

0

0

0

1

1

1

1

1

-

⊥

l

一

l

l

l

l

L]!l!qt?qOJdfiu!qooq.-2^0

1.tI.1.~.i-･.-･-I.Ii-･･･-･

8.0

玄　6.0'=

隻/■■ヽ

云　4.0

2.0

0

0.01　　　　　　0.1　　　　　　1,0　　　　　　　10

To〔ms]

Figure 5: Tも･e relationship between To and Bo

The rate of lユ5er 〔ens Sent by each LU (Bu bit/S) is given by

Bu-a-B,-Bo (1)

Fig一打e 6 shows the relationshiT) between To and But Bll has a maximllm at a certain value of

To. when the rate of comeCt;ons is 100 Mbit/S, each LU can sends 8.6 Gbit/s of心｡r Cells if

wesetTo to i ns.

4　Conclusions

This paper proposed a distrib-ユted trafBc control method for large multi-stage ATM switching

systems. When a newvirtualcircdt is to be added from some lap-ユt uI to some Olユtp11t

LU, a system muS七五nd an a･ppropriate path through the switch networkwith s11氏cient unlBed

bandwidth on each of its hnks to accom皿Odate地･e newvirtual circuit･ h the proposed switching

system, each port of the basic switches has its own tra毘c modtor, and each LU periOdicauy

obtahs congestion information about available paths from OAM cells in order to route a new

virtualcircdt independently. The performance of the proposed system depends on the interval

between OAM cells･ If we set it too long, the congestion information in LUs is delayed, blユ七

if we set it too short, the number of OAM celk in the system increases. We showed how an

appropriate interval can be detemied in order to maximize tも.e number of user cells that each

LU can send. This tra艮c control methodwin s.it future Tbit/s ATM switching systems･

References

l1] K･ Genda and N･ Yamanaka, "TORtJS: Terabit-per-Second ATM Switching System Archi-tecture Based on Distributed hternalSpeed-Up ATM Switch," IEEE Joumalof Selected

Areas in ComunicationS,γol. 15, No. 5, 1997.

1 1 3

0.1　　　　　1.0　　　　　10

To lms]

Figure 6: The relationship between To and Bu

【2】 N. Yamanaka, S. Yasl血wa, E･ Oki and T･ Kawanura, "OPTIMA‥ Tb/s ATM Switching

system ArchitectlEe: Based on Highly StatisticalOpticalWDM hterco-ection,M Proc･IEEE ISS'97, System ArchitectIEe, 1997l

l3] K. Nakai, E･ Oki and N･ Yamanaka, "Perfo-ance Evaluation of ATM Switch ArchitectlEeUsing WDM Grouped Links," Tech･ Report of ⅡICE, SSR-98-49 (1998-07)･

[41 K. Nakai, E･ Oki and N･ Yamanaka, 't'Scalable 3-Stage AT氾 Switch Architecture Using Opti-

CalwDM Grouped Li血s Based on Dyna血c Bandwidth Sha血g,77正ICE TranS･ Co二-i-ln･,

Feゎ,1999,

[5] K_ ShiomOtO and NI Ya皿anaka, "AnAdmission control scheme based on measurement of

ins七antmeOuS utilization,"正=CE Trans･ Comml皿∴, Vol･ J80-B-I, Nol12, pp･ 950-960, Dec･

1997.

【6】 E. Oki and N. Yamanaka, "High-speed connection ad血ssion control in ATM networks by

generatingvirtualrequests for connection7n Proc･ IEEE ATM'98 Workshop, pp･ 295-2997May1998･

[71 N. Yamanah,i(. Shiom0to and H･ flasegawa, "ALPEN: A simple and aexible ATM networkbased on multiprotocol emulation at edge nodes,77IEICE Trans･ Commun, E79-B, No･ 4, pp･

611-615 (1996)･

1 14

l一｢.--