Performance evaluation of dependable real-time communication with elastic QoS

When a client requests a real-time connection that requires an excessive amount of resources and/or a very high level of QoS, the network service provider may have to reject the request, and only a small number of connections could be accepted. On the other hand, if the client requests only the minimum level of QoS, he may receive only a bare-bones service even when there are plenty of resources available. One way of utilizing resources efficiently is to specify flexible (elastic) QoS requirements that can be adapted to the availability of network resources. S.J. Han et al. (1997) proposed to allocate one primary channel and one or more backup channels to each dependable real-time (DR) connection. One drawback of this scheme is the severe reduction in number of DR connections that can be accommodated, due mainly to the need for reserving resources for backups. This is equivalent to wasting precious resources in the absence of faults as far as the system´s ability of accepting DR connections is concerned. By using elastic QoS for this DR communication service, one can accept substantially more DR connections and improve the utilization of resources efficiently and significantly. We analyze the DR communication service with elastic QoS. Fault tolerance is achieved by allocating one backup channel to each DR connection. A Markov model is developed and used to analyze the average QoS level allotted to the primary channel of each DR connection. Our evaluation results show that the proposed Markov model accurately represents the behavior of DR connections with elastic QoS.