PERFORMANCE BOUNDS FOR MODELING NUMA ARCHITECTURES

Scalable, multi-processor, computing systems in which memories are log ically shared but physically distributed exhibit non-uniform memory ac cess (NUMA) times. Modeling such systems presents special difficulties . Closed queuing networks, in which some servers provide exponentially distributed service and others provide deterministic (constant) servi ce, offer the desired level of model representation, but such mixed ne tworks remain analytically intractable. This paper shows that the thro ughput of any such mixed network is necessarily bounded below by the t hroughput of a purely exponential-server network and bounded above by the throughput of a purely deterministic-server network. In particular , replacing any deterministic server by an exponential server of the s ame mean will not increase throughput. Since fast techniques exist for solving purely exponential and purely deterministic networks, perform ance bounds are at hand. (C) 1997 Elsevier Science B.V.