Diskless checkpointing

Diskless Checkpointing is a technique for checkpointing the state of a long-running computation on a distributed system without relying on stable storage. As such, it eliminates the performance bottleneck of traditional checkpointing on distributed systems. In this paper, we motivate diskless checkpointing and present the basic diskless checkpointing scheme along with several variants for improved performance. The performance of the basic scheme and its variants is evaluated on a high-performance network of workstations and compared to traditional disk-based checkpointing. We conclude that diskless checkpointing is a desirable alternative to disk-based checkpointing that can improve the performance of distributed applications in the face of failures.