PERFORMANCE STUDY OF STATE DEPENDENT PRIORITY SCHEMES FOR ATM MULTIPLEXER

Abstract

Congestion control is an important function in ATM network. Since traffic with diverse QOS requirements, competes for ATM network resources , some form of prioritization among the traffics is necessary at cell level. Specially if an aggressive call admission control is followed, cell level prioritisation becomes critically important so as to ensure that different connections receive their negotiated QOS. In this thesis, an optimal Queue length threshold scheduling policy for two types of traffic, viz real time and non real-time is studied. In this policy if the number of non real time traffic exceeds a threshold «, fixed number of servers are allocated to the non real-time traffic, while the remaining servers are scheduled with policy that real time traffic has higher priority over the non real time traffic; otherwise all the servers are scheduled with the above priority policy. The input arrival is assumed to be batch poisson arrival with uniform distribution. A new policy called dynamic optimal queue length threshold policy is proposed, which dynamically increase the number of servers allocated to type 2 traffic depending on type 2 traffic intensity. The above two policies are compared with fixed priority policy, shortest length first policy and longest length first policy under all possible traffic conditions. The simulation is done in C++ and runs on TATA ELXI system.