Bandwidth allocation of multiple QOS classes in ATM environment

Abstract

For future Broadband-ISDN

Tight Upper Bounds for Cell Loss Probabilities and Required Bandwidth Estimation in ATM Multiplexers

“…Accuracy of the required bandwidth estimates In the following, to illustrate the performance of the proposed upper bounds for the e ective bandwidth, some numerical examples are given. Our results are compared to the alternative solutions based on approximations (5), (6) In Table III,we show the required bandwidth estimated by di erent methods, for di erent bu er sizes. The input tra c is a mix of 20 class-1 sources with a CLR requirement equal to 10 ;5 , 50 class-2 sources with a CLR requirement equal to 10 ;7 and 100 class-3 sources with a CLR requirement equal to 10 ;3 .…”

“…In the same spirit, a new approximation is proposed in 8]. The factor a 0 h 0 1i is replaced by a new factor L 0 obtained from a bu erless model by using large deviation theory, leading to G(x) L 0 e z 0 x : (6) L 0 is actually the Cherno 's large deviation approximation to the probability that the instantaneous input rate exceeds the channel capacity 12]. This factor L 0 is very accurate, however, it constitutes an approximation to the over ow probability instead of an upper bound so that when the bu er size is relatively small, this approximation may underestimate the exact over ow probability b y a theoretically unknown amount.…”

A Simulation Study of VP Bandwidth Allocation for Heterogeneous Traffic in ATM Networks

“…Using the formula in [3] to compute the individual cell loss probabilities, an iterative algorithm is proposed below to determine the sufficient and minimum capacity of a VP so that all the individual Ql are satisfied when different traffic classes are multiplexed together on the VP. The algorithm determines C or Om) for a given buffer size K or K(-).…”

“…In the above algorithm, step 3 computes the individual cell loss probabilities u, by using the formula provided in [3]. In [3], a simple and accurate algorithm is proposed to compute Q by applying an MMDP/D/1/K queue [8] for the burst-scale congestion component and an M/ D/ 1/ K queue for the cellscale congestion component when various traffic classes are multiplexed together on a queue; other methods [1,6] can also be employed to compute <7 .…”

“…A VP combination is defined in [3] Another more flexible approach is to provide some priority handling mechanism inside the network. The network can make use of this priority handling capability to achieve different QoS while maintaining efficient network use [4,5].…”