A NEW FAST FAULT TOLERANT SCHEDULING APPROACH IN DISTRIBUTED SYSTEMS

摘要

The general mechanism of providing fault tolerance in distributed scheduling systems is by task redundancy where in a backup task (which is an exact copy of the original task) is created and scheduled for each existing task. This paper follows a new approach to address the task redundancy issue: Instead of generating a separate backup task, the same task is scheduled twice. Also this copy of the task is scheduled on the same time unit (as far as possible) on a different processor. In case where two processors at the same time unit are not available, the task can be scheduled at different time units and after all the tasks are scheduled, the schedule is optimized to see if any of the tasks can be scheduled earlier. The paper presents performance analysis and simulation results which suggest that this approach is advantageous for multiprocessors in general and for even numbered multiprocessors in particular.