Distributed Shared Memory: To Relax or Not to Relax?

摘要

Choosing a memory consistency model is one of the main decisions in designing a Distributed Shared Memory (DSM) system. While Sequential Consistency (SC) provides a simple and intuitive programming model, relaxed consistency models allow memory accesses to be parallelized, improving runtime performance. In this article we compare the performance of two multithreaded memory coherence protocols. The first protocol implements Home-based Lazy Release Consistency (HLRC) memory semantics and the second one implements SC semantics using a MultiView (MV) memory mapping technique. This technique enables fine-grain access to shared memory while using the virtual memory hardware to track memory accesses. We perform an "apple-to-apple" comparison on the same testbed environment and benchmark suite, and investigate the effectiveness and scalability of both these protocols.