Multiprocessor cache analysis using ATUM

The design of high-performance multiprocessor systems necessitates a careful analysis of the memory system performance of parallel programs. Lacking multiprocessor address traces, previous multiprocessor performance studies using analytical models had to make an inordinate number of assumptions about the underlying memory reference patterns. We previously developed a scheme called ATUM - Address Tracing Using Microcode - to get reliable operating system and multiprogramming traces on single processors. This paper briefly describes the multiprocessor extension of ATUM and its implementation on a VAX 8350 multiprocessor. We also report on our use of the resulting traces to analyze physical versus virtual addressing of large caches, process-identifier hashing in virtual caches, cache interference between multiple processes, cache interference between multiple CPUs, process affinity, and semaphore usage in writeback caches.