NWcache: Optimizing disk accesses via an optical network/write cache hybrid

Abstract

In this paper we propose a simple extension to the I/O architecture of scalable multiprocessors that optimizes page swap-outs significantly. More specifically, we propose the use of an optical ring network for I/O operations that not only transfers swapped-out pages between the local memories and the disks, but also acts as a system-wide write cache. In order to evaluate our proposal, we use detailed execution-driven simulations of several out-of-core parallel applications running on an 8-node scalable multiprocessor. Our results demonstrate that the NWCache provides consistent performance improvements, coming mostly from faster page swap-outs, victim caching, and reduced contention. Based on these results, our main conclusion is that the NWCache is highly efficient for most out-of-core parallel applications.