QUALITY OF SERVICE MANAGEMENT IN ATM NETWORKS

Abstract

Broadband applications over ATM networks generate traffic with varying traffic characteris tics. These applications specify the required service guarantees in terms of quality of service (QoS) that the network has to provide. Several elements are needed to provide QoS guarantees. Examples are QoS specification, Admission control, QoS negotiation, Resource allocation and scheduling and Traffic policing. ATM networks should also support QoS adaptation and QoS renegotiation techniques. To ensure that all such requirements are satisfied, quality of service management is essential in ATM networks. The work presented in this thesis is an effort to study and propose different mechanisms for the management of quality of service in ATM networks. In the first part, we propose amechanism for QoS provisioning through dynamic bandwidth allocation and buffer control. Satisfaction of the different QoS requirements is aresource al location problem. To make efficient use of resources and to satisfy QoS requirements like CLR or delay, efficient bandwidth and buffer allocation methods are needed. We assume that a large pool of buffer space exists that releases or receives back the buffer space as per requests received for buffer allocation. Under the proposed scheme the allocated buffer space of the connection is reallocated based on measurement of data from the system for certain time du ration. Our scheme is based on the fact that there exists arelationship between queuing buffer size, cell losses and delay. This scheme provides optimal bandwidth and buffer allocation to satisfy multiple QoS namely cell loss rate (CLR) and delay. We also investigate the use of cell loss priority (CLP) bitof cell header. In the second part, we propose an algorithm for pre-computation of QoS routes. Path precomputation schemes benefit from having multiple candidate routes to each destination, to balance the network load and have additional routing choices in case of aset-up failure. The QoS constraints for traffic are interrelated in away that is determined by the network scheduling discipline. We have assumed that the scheduling policy in the network is rate based. We compare the performance of our algorithm with that of pre-computation of /c-constrained QoS routes using modified Bellman-Ford algorithm and with on-demand routing to compute route to the destination. We performed the simulation studies using minimum hop, widestshortest path and shortest-widest path optimality criteria on two different topologies namely ISP and Switched cluster. We assume that aconnection blocks after all kconstrained paths are exhausted and result in connection setup failure. In our simulation after afailure in route extraction or connection establishment, the source triggers re-computation of the kconstrained paths. We also carried out an experiment where the pre-computation rate was fixed and the link-state update period was varied. In the third part we present anew adaptation protocol for QoS adaptation with renegotiation that allows an ATM network to recover from the QoS violations in order to satisfy end-to-end QoS requirements. Our protocol is applicable to PNNI based Am network where the nodes are grouped together in peer groups hierarchically. We assume that every node has aQoS/Route Monitor unit that receives QoS/LSU updates from the network on aperiodic or triggered basis. In addition to its usual functions, this monitor would also function as QoS agent, QoS manager or Connection QoS Manager depending on its location. The QoS/Route Monitor is responsible vi for sending or receiving the signalling messages required in our protocol. There are two categories of QoS parameters that have been considered - concave and ad ditive. The QoS violation may occur in both types of QoS parameters. The protocol works differently for each of these categories. To facilitate the functioning of the protocol, several signalling messages have been defined. These signalling messages characterize different con trol mechanisms required for the protocol and the network. Finally, we conclude with a unified model for quality of service management in ATM net works that has QoS monitoras primarypart of theATM interface or switch. The modelconsists of components responsible for QoS provisioning, QoS monitoring and managing QoS routes. The model is also responsible for processing the proposed adaptation protocol between end points.