Research Initiation Award: Fast Checkpointing in Parallel Environments

点击翻译按钮获取中文摘要

9409496 Plank The main goal of this project is the evaluate several techniques for providing efficient check-pointing in parallel processing systems. Check-pointing is the act of saving the execution state of a program with the purpose of restoring the system to a stable state after a failure. The major obstacle to providing general purpose fault-tolerance for parallel programs is the overhead incurred by check-pointing. In an attempt to improve the performance of check-pointing, this project explores diskless check-pointing, user controlled check-pointing and meshing of check-pointing with log-structured file systems. The impact of these approaches and their subsequent implementations will be investigated. ***