The Analysis and Comparison for the Software Cache Coherence Schemes

摘要

In a memory hierarchy for a multiprocessor system, data inconsistency may occur between adjacent levels or within the same level. Using private data caches can reduce network traffic and average memory access latency, but they will introduce a cache coherence problem since multiple copies of a memory location can be resident in several different caches simultaneously. Several software cache coherence schemes were proposed. However, the damage of cache coherence in a Doacross loop is more serious than Doall loop. In order to evaluate the impact of various cache coherence schemes in the Doacross loop, several simulators are developed to analyze the results. Experimental results show that TPI scheme has the better performance.