No abstract

No abstract

No abstract

FaCSimilC(02x92 3838 ABSTRACT For badband links urrying ATM traf&, a model is -sed this paper that enables useful mta~ures of link paformance to be obtained quite d k d y , for both delay sensitive and loss sensitive smriw. It is assumed that the haEc mix on the l i n l r r d t s in aunifom d v a l p r o c e~~ for cells, and that the link cqmcity is suf€icientiy gnatthnt burstiness m individual services not i n v~t e the rmiformiry assumption. From the modcl (which we term the p' model) apasimirticbound for the waiting rimedistibutionof avirmal cell is daived. Usingthis result, simple formulas m found forestirnoting the buffer size required to achieve a given probability of loss (e.g. I O -' ' ) . Keywonis ATM, SONET. B-ISDN. qwueing model, loss sensitive ua€Iic, delay sensitive trpffic. buffer size. p W t y of loss. L Inlrodnci3onThe h e f i t to d c e users and providers dike, arising from rhc pvision of a single integrated communicafjm neework to clary voice. data and video m€fk, @acing a plethora of dedicated nemxks has long been acknowledged [I].Continuing developmem. panicukly in fibre optics are making the wide mmnisii bandwidth required by such a system incrrasingly auainable [VI.Asynchronous Transfer Mode (ATM) has been widely recognised (CClTT Recommmdation 1121) as a promising way of arranging for information amsf= rhmugh this systan Mned the Broadbend Integrated Services Digital Network (B-ISDN) [4]. The B-ISDN is expected to cauy services with average bit rate &cm a few kbh to several hundred Mb/s. Some services produce cells in acodnuou stream whereas 0th-produce cells in aninmmitrmtorbumty farhion adsnnesavicespmduce h i & correlated cell streams [a]. Funha, savices canbe dauificd inro lars msitive and delcy sensitive classes. To evaluate rhepafonnmce of a broadband WS aknowledge of the anivalpcess is r e q u i d If M c arrives at ono& mnindqm&u srredms, each sueam having a wnstant inter-miVal time, a m c a t i r m of the nD/D/l/K model r.81. allowing forthe differing arrival rates between streams, could be used to determine the exact buff= Size Kquired and the link could be NII without loss.In pr~ticehowever, the auivalprocss will not be this pndictable and a service urn only &gin at the start of a time slot 0.). The hUD/1 model, although not mmporatiug priority classes explicitly or a slot effect. has been used to study link performance and is claimed to be quite robust to mf6c mix variation VI. Orher models used have included elements to repnsenr explicitly, bursty and continuous components of the haffic mix [9,10.11,1213]. R d suggests that the need for this Complur approach dimhkhs when the maximum service rate is less thm~ h t 10% or so of the linkrate [14].Fortroadbandlinks, amerhod isprop~sedin this paper, btenabk u s e € i a l~u r e s of linkpafarmrmce to be obtained quite diredy. Usmg this method, a set of simple formulas is daiv.dforestimating the buffer size mpbd to achieve agivm probability of loss, a pessimistic b o d on the waiting time distribution of cells. and the disniition of qu...