Slow memory: the rising cost of optimism

Abstract

Rapid progress in the design of fast CPU chips has outstripped progress in memory and cache performance. Optimistic algorithms would seem to be more vulnerable to poor memory performance because they require extra memory for state saving and anti-messages. We examine the performance of both optimistic and conservative protocols in controlled experiments to evaluate the effects of memory speed and cache size, using a variety of applications.

An Evolutionary Algorithm to Optimize Log/Restore Operations within Optimistic Simulation Platforms

“…As pointed out in the Introduction, beyond the above cited direct costs associated with the adopted log/restore parameters, we may also have hidden effects, such as locality effects (see, e.g., [19]) and the so-called thrashing phenomena. The latter is related to the variation of the latency of the state restore procedure (e.g., due to variations of the coastingforward latency after a re-tune of the log interval), which can produce cross-object interactions negatively affecting performance.…”

An overhead reducing technique for Time Warp

“…We know that, ts(e next ) ≥ ts(e 0 ') ≥ ts(e'') ≥ lbt τ0' (P'). Therefore, lbt(e 0 ) = lbt τ0' (P') + δ Consider, lbt(e), lbt(e) ≥ lbt τ' (P') + δ ≥ lbt τ0' (P') + δ = lbt(e 0 ) = lbt τ (C) --- (11) From (10) and (11), the claim is true for case 2B. Therefore the claim is true for any normal event e. (13), (14) and (15), the claim is true for an antimessage.…”