Comparative evaluation of latency-tolerating and -reducing techniques for hardware-only and software-only directory protocols

We study in this paper how effective latency-tolerating and -reducing techn iques are at cutting the memory access times for shared-memory multiprocess ors with directory cache protocols managed by hardware and software. A crit ical issue for the relative efficiency is how many protocol operations such techniques trigger. This paper presents a framework that makes it possible to reason about the expected relative efficiency of a latency-tolerating o r -reducing technique by focusing on whether the technique increases, decre ases, or does not change the number of protocol operations at the memory mo dule. Since software-only directory protocols handle these operations in so ftware they will perform relatively worse unless the technique reduces the number of protocol operations. Our experimental results from detailed archi tectural simulations driven by six applications from the SPLASH-2 parallel program suite confirm this expectation, We find that while prefetching perf orms relatively worse on software-only directory protocols due to useless p refetches, there are examples of protocol optimizations, e.g., optimization s For migratory data, that do relatively better on software-only directory protocols. Overall, this study shows that latency-tolerating techniques mus t be more carefully selected for software-centric than for hardware-centric implementations of distributed shared-memory systems. (C) 2000 Academic Pr ess.