Maximizing Job Completions Online

We consider the problem of maximizingthe number of jobs completed by their deadlinein an online single processor system where the jobs are preemptable and have release times. So in the standard three eld scheduling notation, this is the online version of the problem 1 j ri pmtn j P(1;Ui). We present a deterministic algorithm Lax, and show that for every instance I, it is the case that either Lax, or the well-known deterministic algorithm SRPT (Shortest Remaining Processing Time), is constant competitive on I. An immediate consequence of this result is a constant competitive randomized algorithm for this problem. It is known that no constant competitive deterministic algorithm exists for this problem. This is the rst time that this phenomenon, the randomized competitive ratio is constant in spite of the fact that the deterministic competitive ratio is nonconstant, has been demonstrated to occur in a natural online problem. This result is also a rst step toward determining how an online...

• Type: Conference
• Year: 1998
• Where: ESA
• Authors: Bala Kalyanasundaram, Kirk Pruhs
• Keywords: Algorithms, Competitive, Deterministic Algorithm, Deterministic Algorithm Lax, ESA 1998