On-line scheduling to maximize task completions

The problem of uniprocessor scheduling under conditions of overload is investigated. The system objective is to maximize the number of tasks that complete by their deadlines. For this performance metric it is shown that, in general, any on-line algorithm may perform arbitrarily poorly as compared to a clairvoyant scheduler. Restricted instances of the general problem for which on-line schedulers ran provide a guaranteed level of performance are identified, and on-line algorithms presented for these special cases.<
>