Scheduling Parallel Processes Without a Common Scheduler.

摘要

An algorithm which solves the critical section problem for distributed processes is presented. We extend the solution of Lamport LL76 by continuing to allow processes to access their respective critical sections in any arbitrary user-specified order, but with greatly reduced storage requirements for each process. In addition, we supply a facility for testing the presence of deadlock among processes waiting to enter their critical code. We show our scheme to be tolerant of several malfunctioning processors and derive an equation relating the probability of total system failure to the probability of many individual failures occurring simultaneously among the processors. (Author)