Delaying Coherence Requests to Enhance the Performance of Strict Consistency Models

Advances in ILP techniques enable strict consistency models to relax memory order through speculative execution of memory operations. However, ordering constraints still hinder the performance because speculatively executed operations cannot be committed out of program order for the possibility of mis-speculation. In this paper, we propose a new technique which allows memory operations to be non-speculatively committed out of order without violating consistency constraints. Consistency constraints are guaranteed through delaying the coherence requests. The proposed technique also improves the performance of spin lock primitives such as TTS lock or MCS lock. Through delaying early acquire requests, the lock transfer time can be improved when there is high contention for a lock. key words: multiprocessor, distributed shared memory, memory consistency model, ILP