Exploring Memory Access Similarity to Improve Irregular Application Performance for Distributed Hybrid Memory Systems

Abstract

With the increasing problem complexity, more irregular applications are deployed on high-performance clusters due to the parallel working paradigm, and yield irregular memory access behaviors across nodes. However, the irregularity of memory access behaviors is not comprehensively studied, which results in low utilization of the integrated hybrid memory system compositing of stacked DRAM and off-chip DRAM. To address this problem, we devise a novel method called <i>Similarity-Managed Hybrid Memory System</i> ( <i>SM-HMS</i> ) to improve the hybrid memory system performance by leveraging the memory access similarity among nodes in a cluster. Within <i>SM-HMS</i> , two techniques are proposed, <i>Memory Access Similarity Measuring</i> and <i>Similarity-based Memory Access Behavior Sharing</i> . To quantify the memory access similarity, memory access behaviors of each node are vectorized, and the distance between two vectors is used as the memory access similarity. The calculated memory access similarity is used to share memory access behaviors precisely across nodes. With the shared memory access behaviors, <i>SM-HMS</i> divides the stacked DRAM into two sections, the <i>sliding window section</i> and the <i>outlier section</i> . The shared memory access behaviors guide the replacement of the <i>sliding window section</i> while the <i>outlier section</i> is managed in the LRU manner. Our evaluation results with a set of irregular applications on various clusters consisting of up to 256 nodes have shown that <i>SM-HMS</i> outperforms the state-of-the-art approaches, <i>Cameo</i> , <i>Chameleon</i> , and <i>Hyrbid2</i> , on job finish time reduction by up to <inline-formula><tex-math notation="LaTeX">$58.6\%$</tex-math></inline-formula> , <inline-formula><tex-math notation="LaTeX">$56.7\%$</tex-math></inline-formula> , and <inline-formula><tex-math notation="LaTeX">$31.3\%$</tex-math></inline-formula> , with <inline-formula><tex-math notation="LaTeX">$46.1\%$</tex-math></inline-formula> , <inline-formula><tex-math notation="LaTeX">$41.6\%$</tex-math></inline-formula> , and <inline-formula><tex-math notation="LaTeX">$19.3\%$</tex-math></inline-formula> on average, respectively. <i>SM-HMS</i> can also achieve up to <inline-formula><tex-math notation="LaTeX">$98.6\%$</tex-math></inline-formula> ( <inline-formula><tex-math notation="LaTeX">$91.9\%$</tex-math></inline-formula> on average) of the ideal hybrid memory system performance.