FAULT-TOLERANT LPT TASK-SCHEDULING IN MULTIPROCESSOR SYSTEMS

Minimizing the finish time of a set of independent tasks executed on a muitiprocessor system, where no faults occur, is a well-known NP-complete problem. However, an approximated schedule can be found efficiently by means of the so called largest processing time (LPT) rule. This scheduling problem when extended to faulty environments may seem daunting. However, this paper shows how to extend the LPT rule to a fault-tolerant non-uniform memory access (NUMA) multiprocessor employing a primary copy and a backup copy for each task in order to handle processor failures. The concept of expected schedule length is also introduced to evaluate the performance of the resulting schedule in case of processor failures. Simulation studies are used to reveal interesting trade-offs associated with the scheduling algorithm.