Overload tolerance for single-processor workloads

In firm real time environments in which tasks must complete by their deadlines if they are to be of any value to the system, it is known that no uniprocessor online scheduling algorithm can guarantee to perform particularly well under conditions of overload as compared to clairvoyant algorithms. The article explores the issue of designing online scheduling algorithms that use multiple processors to compensate for their lack of clairvoyance. In particular, it is shown that given enough processors, online scheduling algorithms can be designed with performance guarantees arbitrarily close to that of optimal clairvoyant uniprocessor scheduling algorithms.