Exploring the limits of disjoint access parallelism
Amitabha Roy, Steven J. Hand, Tim Harris · 2009
The desire to extract parallelism has recently focused attention on improving concurrency control mechanisms in applications. However, programmers have few tools to help them understand the synchronisation bottlenecks that exist in their programs – for example to identify which locks are heavily contended, and whether different operations protected by the same lock are likely to conflict in practice. This kind of profiling is valuable to guide programmers in efforts towards optimising their lock-based concurrent applications, as well as to aid the design and tuning of advanced concurrency mechanisms such as hardware and software transactional memory systems. To this end, we describe the design and architecture of a prototype tool that provides insights about critical sections. Our tool is based on the Pin binary rewriting engine and works on unmodified x86 binaries. It considers both the amount of contention for a particular lock as well as the potential amount of disjoint access parallelism. We present the results of applying the tool to a representative sample of applications and benchmarks. 1