Two-Level Failure Recovery Schemes for Multicomputers and Distributed Systems

点击翻译按钮获取中文摘要

The thrust of this proposal is on backward recovery schemes for multicomputer and distributed systems. The proposed research is based on a two-level approach that uses a low overhead recovery scheme to tolerate more probable failures, while the less probable failures may be tolerated with a higher overhead. By minimizing the overhead for the more frequently occurring failure scenarios, the proposed two-level approach will achieve lower average performance overheads as compared to existing recovery schemes. Objectives of the proposed research are 1) the design of efficient two-level recovery schemes, 2) the development of optimization techniques for two-level recovery schemes to determine various parameters (e.g., checkpointing frequency) that willminimize the average performance overhead fo the recovery scheme, 3) the implementation of two level recovery schemes on a 64 processor nCUBE-2 multicomputer and on a network of workstations, and 4) the experimental evaluation of two-level recovery schemes to demonstrate their ability to achieve better performance than the existing recovery schemes. The goal of the proposed education activities is to incorporate the results of recent research into undergraduate and graduate curriculum. specific objectives fo the education plan are 1) the development of an undergraduate introductory course on fault tolerance and a graduate course on fault tolerance techniques for parallel and distrubuted systems, 2) the development of an associated undergraduate research effort, and 3) the development of a software toolkit to facilitate experimetns in the proposed courses.