RUN-AHEAD CRASH RECOVERY FOR SHARED VIRTUAL MEMORY SYSTEMS

摘要

Shared virtual memory (SVM) is an attractive approach for providing a simple parallel programming environment on a cluster of computers since it permits programmers to assume the existence of shared memory image across physically distributed-memory systems, obviating the need of explicit message passing operations. While several checkpointing and message logging techniques for SVM have been proposed and evaluated extensively, very little attention has been given to crash recovery issues. In this paper, we propose and evaluate a new crash recovery technique, called run-ahead crash recovery (RCR). Our experimental results confirmed that RCR provides a swift crash recovery support to SVM and outperforms a simple crash recovery by 2% to 67% under seven parallel applications examined.