Xunify-a performance debugger for a distributed shared memory system

Although distributed shared memory (DSM) provides a convenient programming model for parallel applications running on a network of workstations, the DSM model presents new challenges for performance evaluation and debugging tools. Although processes communicate and cooperate via shared memory and synchronization primitives, most performance problems arise from the high cost of implementing these abstractions with network messages. Because the DSM abstraction hides the underlying message passing, it is often difficult to identify the reason for poor performance. We have developed an instrumentation system and a performance evaluation tool called Xunify for monitoring and performance debugging of a DSM system. Xunify provides information in terms of the programming model level, logical message level, and the transport protocol level, helping program writers quickly pinpoint the source of inefficiency