Performance modelling and evaluation for the XMP shared-bus multiprocessor architecture

Abstract

This paper presents the performance modelling and evaluation of a shared bus multiprocessor, XMP. A key characteristic of XMP is that it employs a special shared bus scheme featuring separate address bus and data bus with split transaction, pipelined cycle (called SSTP scheme). To assist evaluating the architectural alternatives of XMP, the features of the SSTP bus scheme as well as two important performance impacting factors: (1) cache, bus, and memory interferences and (2) DMA transfer, are modelled. We employ a Subsystem Access Time (SAT) modelling methodology. It is based on a Subsystem Access Time Per Instruction (SATPI) concept, in which we treat major components other than processors (e.g. off-chip cache, bus, memory, I/O) as subsystems and model for each of them the mean access time per instruction from each processor. Validated by statistical simulations, the performance model is fed with a given set of representative workload parameters, and then used to conduct performance evaluation for some initial system design issues. Furthermore, the SATPIs of the subsystems are directly utilized to identify the bottleneck subsystems and to help analyze the cause of the bottleneck.