Issues in reliable network memory paging

Recent distributed systems are connected with high-performance networks that make possible the use of a new level of memory hierarchy: network memory. Network memory provides both high bandwidth and low latency, that make it attractive for uses like paging and file caching. In this paper we explore the issues of building a reliable network memory system that is resilient to single workstation failures. We propose novel parity-based policies and evaluate their performance using trace-driven simulation of realistic applications. The presented performance results suggest that our proposed policies provide reliability at a surprisingly small run-time overhead