Fault-tolerance has gained renewed importance with the proliferation of high-performance clusters. However, fault-tolerant systems have not yet been widely adopted commercially because they are either hard to deploy, hard to use, hard to manage, hard to maintain, or hard to justify. We have developed M 3, a practical and easily-deployable multiple fault-tolerant MPI system for Myrinet, to satisfy the demand for a fault-tolerant system.
In this paper, we run rigorous tests using real-world applications to validate that M 3 can be used in commercial clusters. We also describe improvements made to our system to solve various problems that arose when deploying it on a commercial cluster.
This paper models our system’s checkpoint overhead and presents the results of a series of tests using computation- and communication-intensive MPI applications used commercially in various fields of science. The experimental results show that not only does our system conform to various types of running environment well, but that it can also be practically deployed in commercial clusters.