Components and Interfaces of a Process Management System for Parallel Programs

摘要

Parallel jobs are different from sequential jobs and require a different typeof process management. We present here a process management system for parallelprograms such as those written using MPI. A primary goal of the system, whichwe call MPD (for multipurpose daemon), is to be scalable. By this we mean thatstartup of interactive parallel jobs comprising thousands of processes isquick, that signals can be quickly delivered to processes, and that stdin,stdout, and stderr are managed intuitively. Our primary target is parallelmachines made up of clusters of SMPs, but the system is also useful in moretightly integrated environments. We describe how MPD enables much fasterstartup and better runtime management of parallel jobs. We show how closecontrol of stdio can support the easy implementation of a number of convenientsystem utilities, even a parallel debugger. We describe a simple but generalinterface that can be used to separate any process manager from a parallellibrary, which we use to keep MPD separate from MPICH.