PTSE RATES IN PNNI NETWORKS Norert MERSCH 1 Siemens AG, Hofmnnstr. 51, D-81359 Münhen, Germny Peter JOCHER 2 LKN, Tehnishe Universität Münhen, Arisstr. 21, D-80290 Münhen, Germny Lrs BURGSTAHLER 3 IND, Universität Stuttgrt, Pfffenwlring 47, D-70569 Stuttgrt, Germny Astrt: The Privte Network Noe Interfe (PNNI), stnrize y the ATM Forum (see [1]), provies flexile n slele routing rhiteture for ATM networks omprising routing protool n signling protool. In PNNI networks PTSEs re use to istriute topology informtion through the network. This pper investigtes the PTSE rte in PNNI noes ue to flooing. Besie theoretil results, the pper lso presents some mesurements of exmple networks. 1 Introution The Privte Network Noe Interfe (PNNI) efine y the ATM-Forum (see [1]) provies flexile n slele routing rhiteture for ATM networks omprising routing protool n signling protool. PNNI routing inlues mehnisms for the utonomous exhnge of ggregte topology informtion to form hierrhil representtion of the network. PNNI signling is se on suset of UNI 4.0 signling. To investigte the hrteristis of PNNI networks two powerful n verstile tools, PNNI emultor n PNNI simultor, hve een evelope. While erlier investigtions (see [2] n [3]) onentrte on the exploittion of the network pity y PNNI routing mehnisms, this pper is fouse on performne spets of the PNNI routing protool. 2 PNNI Topology Informtion Distriution 2.1 Topology Informtion PNNI uses soure routing to etermine pth through network. Hene, every noe nees omplete esription of the topology to perform the neessry omputtions. The PNNI protool provies three levele t struture for topology informtion. On the first level re the informtion groups (IG). Eh IG only overs one speifi prt of noe, 1 KEY CONTACT: norert.mersh@in.siemens.e; Phone +49-(0)89-72224207; Fx +49-(0)89-72222090 2 joher@ei.tum.e; Phone +49-(0)89-28923506; Fx +49-(0)89-28923523 3 urgsthler@in.uni-stuttgrt.e; Phone +49-(0)711-6857966
e.g. one port n its resoures. On seon level, the IGs re unle in PNNI Topology Stte Elements (PTSEs). PTSEs re the units of flooing n retrnsmission. As they o not ontin informtion out the originting noe, they nee some kin of envelope when eing sent to neighor. The PNNI Topology Stte Pket (PTSP) is suh n envelope tht trnsports PTSEs inluing informtion out the noe s ientity. 2.2 Flooing Flooing is relile metho to istriute informtion within network. Its min vntge, ut lso its min rwk is reunny. There re two mjor resons, why noe originlly floos PTSE: Triggere Upte: Triggere flooing hppens if ompletely new PTSE is originte y noe or if there is signifint hnge in n informtion group within n existing PTSE. Re-Origintion (Refresh): To prevent the PTSE from eing elete from the topology tse the originting noe re-origintes PTSE perioilly, even if the ontents i not hnge. On reeipt of PTSE tht is not yet in its tse, noe forwrs this PTSE to ll neighors, exept the one the PTSE hs een reeive from. (1) (2) (3) Figure 1: Exmple of the simple flooing Mehnism Figure 1 shows the flooing of n informtion element originte y noe in n exmple network. In this exmple five informtion elements hve to e proesse y the network. Only three of them woul hve een suffiient to upte the tses of the respetive noes. If j i is the numer of jent noes of noe i, every PTSE originte y noe i must e sent: j i times reeive: 0 times n every PTSE not originte y noe i must e sent: reeive: j i 1 times j i S times S refers to the ft tht, ue to the nture of flooing, prtiulr PTSE is not flooe upstrem on those links, whih form the shortest pth tree (SPT) with the originting noe s the root. Thus the vlue of S epens on the onsiere originting noe n reeiving noe. 2
3 Clulting the PTSE Rtes per PNNI Noe 3.1 Flooing of Aress PTSEs The min ftors influening the ress PTSE rte in PNNI noe re: Numer of ress PTSEs in the network: Time etween the refresh flooing of PTSEs: Numer of neighors of the prtiulr noe: The verge vlue of S for tht noe: NoArPTSE PTSERefreshIntervl NoNeighors AverS Hene the verge rte for reeive ress PTSEs on prtiulr noe (AverAressPTSERte) n e lulte s follows (the reltion refers to the ft tht some of the PTSEs re originte y the noe itself): NoArPTSE ( NoNeighors AverS) AverAressPTSERte (1) PTSE Re freshintervl Within PNNI networks ress summriztion is pplie to reue the numer of PTSEs to e flooe. Hene the ress struture, i.e. the ssoition of resses to PNNI swithes hs very ig influene on the numer of ress PTSEs. Moving resses within network might eteriorte the summriztion of resses within peer groups, resulting in itionl PTSE trffi. 3.2 Flooing of Link Stte PTSEs Conerning the plnning of the routing protool proessing pity it is iffiult to estimte the mximum rte of signifint hnges n thus the flooing rte, sine this hevily epens on the ynmi ehvior of the network n on the PNNI prmeters. As n upper oun only the minimum intervl etween the flooing of PTSE (MinPTSEIntervl) n e tken. This results in the following formul for the mximum rte of reeive link stte PTSEs (MxLsPTSERte) in noe within PNNI network onsisting of l links (gin, the < reltion refers to the ft tht some of the PTSEs re originte y the noe itself): MxLsPTSERte NoNeighors - AverS < l 2 (2) MinPTSEIntervl one for eh iretion Extensive simultion is neee to evlute the ifferent influene ftors of the upte rte n to fin sensile upper oun (or inrese the MinPTSEIntervl). 3.3 Mesurements The mesurements hve een one with the PNNI simultor, whih is very flexile tool to investigte topologies of up to 100 noes. As the PNNI protool is moele in very etile 3
wy, it llows the mesurement of vlues iretly relte to the protool (like PTSE rte, reunny of PTSEs et), s well s rnkk n ll rejetion proility, pylo n signling oupny of the links n ports. Both tres n sttistis re ville for most of the vlues. The network we hve simulte hs squre meshe topology with 16 noes. All links re i-iretionl n hve pity of 155.52 Mit/s (STM-1). Per noe 1 ress PTSE ws generte ontining severl resses. Hene there re 16 ress PTSEs n 48 link stte PTSEs in the network. The PNNI speifi prmeters re set to vlues reommene in the nnex of [1]. The prmeters for the three ifferent trffi lsses re shown in tle 1. Cll rrivl n urtion re moele using Poisson proess. Trffi Clss A B C Bnwith in MBit/se 0.8 4.0 12.0 Cll rrivl (men) in se. (for 10% lo) 72.72 755.28 6976.80 Cll urtion (men) in se. 480 480 480 Tle 1: Soure prmeters In figure 2 the men n the 95% perent quntiles of the PTSE rte re shown for ifferent lo levels, ompring the PTSE rte in enter noe versus orner noe. Figure 2: PTSE rte of enter noe versus orner noe. As expete the PTSE lo in enter noe is higher thn in orner noe, euse: A enter noe hs more neighors thn orner noe. 4
The position of enter noe within the shortest pth tree of flooe PTSE is ifferent from the position of orner noe. This influenes S. In this exmple the PTSE trffi is minly etermine y the numer of triggere uptes of link stte PTSEs, whih inreses with the offere lo. The refresh rte (offere lo = 0) is quite low ue to the smll totl numer of PTSEs (PTSERefreshIntervl = 1800 se oring to [1]). 4 Conlusion In this pper performne spets of the PNNI routing protool hve een presente fousing on the PNNI topology informtion elements n their flooing mehnism. The min ftors etermining the PTSE rte within PNNI network re: the ress struture (ssoition of resses to swithes), the rte of signifint hnges on the links, the meshing of the network, the vlues of the vrious timers n threshols efine in [1], Annex E. An estimtion of the routing protool proessing pity s performe ove yiels only lower oun for the neessry performne pity of PNNI noe, ue to the following resons: PNNI timers re jittere ut flooing is still ursty. PTSE retrnsmissions ue to it errors must e tken into ount. Reserve pity is neee for tse synhroniztion (see [1]). In multiple peer group environment logil group noes emn for itionl pity. Future work will onentrte on further mesurements onerning PTSE performne spets in PNNI networks, inluing the urstiness of the PTSE trffi. Aitionlly, moifie flooing mehnisms will e investigte n evlute (e.g.: [4]). Referenes [1] ATM Forum Tehnil Committee, Privte Network-Network Interfe Speifition Version 1.0 (PNNI 1.0). ATM Forum f-pnni-0055.000, Mrh 1996. [2] U. Gremmelmier, P. Joher, J. Püshner n M. Winter, Performne Evlution of the PNNI Routing Protool using n Emultion Tool. 16 th Int. Swithing Symposium, Toronto, Septemer 1997. [3] P. Joher, J. Frings, U. Gremmelmier n M. Winter, Plnning Aspets of ATM Networks using the PNNI Routing Protool. NOC 98, Mnhester, June 1998 [4] Y. Hung, P. K. MKinley, Swith-Aie Flooing Opertions in ATM Networks. IEEE INFOCOM `97, Koe, April 1997. 5