Call admission control in generalized processor sharing (GPS) schedulers using non-rate proportional weighting of sessions

Abstract

Absrmcr-Generslized Processor Sharing (GPS) is an ideal fluid schedul-ing discipline that supports well defined delay and loss bounds on Leakybucket constrained trafec Its packetized versions (WFQ, WFZQ, PGPS etc...) are considered as the packet scheduler of choice in IP routers and ATM switches of the future. The currently accepted approach for the design of GPS schedulers is based on determiaistic QoS guarantees, which is overly conservative due to the applied loose bounds and leads to Umitations on capacity. We developed a framework for the computation of tighter delay bounds, bandwidth and delay de-coupling ln GPS systems. In this paper, we propose several effective call admission control (CAC) algorithms that work in the bandwidth and delay decoupled system while using the tighter delay bounds presented hereln. One of the proposed CAC algorithm also handles the best-effort service class bestde the QoS guaranteed service classes. Performance evaluation of several CAC algorithms are presented herein.

Optimal Call Admission Control on a Single Link With a GPS Scheduler

“…An acceptable φ assignment is one which is feasible (that is N i=1 φ i < 1 5 ) and delivers the required QoS to 3 Notice that e i refers to the backlog clearing time of session i while b j refers to the j th of the ordered backlog clearing times.…”

“…Its effectiveness and capabilities in guaranteeing a certain level of Quality of Service (QoS) to the supported streams in both a stochastic ( [5], [6], [4]) and deterministic ( [1], [2], [3], [7]) sense have been investigated. Traffic management based on deterministic guarantees is expected to lead to lower network resource utilization compared to that under stochastic guarantees.…”

“…In [14] the above problem is addressed for leaky bucket regulated connections and statistical delay guarantees. The work most relevant to ours is that of [3], where tight delay bounds (also reported in [10]) have been presented in conjunction with a CAC algorithm for the single node case, which aims to calculate a weight assignment for the sessions that would not result in an over-achievement of session delays. In order to do so, the dependencies among the sessions must be considered.…”

“…In order to do so, the dependencies among the sessions must be considered. The CAC procedure in [3] does not address this problem directly since it employs an exhaustive search, having performance bound calculations as an intermediate step. More specifically, the maximum delay experienced by the sessions is determined for a weight assignment and the assignment is modified trying to maximize an objective function.…”

“…More specifically, the maximum delay experienced by the sessions is determined for a weight assignment and the assignment is modified trying to maximize an objective function. While the search in [3] terminates after a finite number of steps, it does not guarantee that an acceptable assignment does not exist if it is not found. This could result in an over allocation of bandwidth and sometimes a call block, even if the bandwidth necessary to guarantee the call's QoS requirements is available.…”