Factor-Based C-AMAT Analysis for Memory Optimization

Abstract

The “memory problem” promotes researches on improving performance of memory systems, as well as researches on proposing more accurate memory metrics. C-AMAT, an extension of AMAT that takes memory concurrency into consideration, can evaluate the performance of modern memory systems more accurately. However, compared to AMAT, the method for calculating C-AMAT is more complicated, besides, additional detecting logic and registers are required to measure parameters of C-AMAT, which incur high hardware overhead for this metric. In this paper, we propose Factor-Based C-AMAT (FC-AMAT), an analysis model based on C-AMAT. FC-AMAT divides a memory system into factors according to actual research demands, and uses factor’s-first C-AMAT to evaluate effects of optimizations applied to the memory system. By selecting factor’s C-AMAT, FC-AMAT can reduce the hardware overhead for measuring its parameters, meanwhile, it guarantees an acceptable evaluation accuracy through a rigorous check. Simulations with varied cache configurations were conducted to verify the usefulness of FC-AMAT. Experimental results show that FC-AMAT can simplify the detecting logic and reduce the storage cost for recording memory access phases, without sacrificing obvious evaluation accuracy, demonstrating the effectiveness of FC-AMAT.