A 0(1) Time Deadlock Detection Scheme in a Single Unit and Single Request Multiprocessor System

In this paper, we propose a new deadlock detection scheme, which is more efficient than existing ones. We assume a special case system where each type of resource has one unit and each request is Iimitecl to one unit request at a time. Unike the previous deadlock detection schemes, our new method takes 0(1) time for detecting deadlock, and O(n+m) time for handling reeource reIease, where n end m are number of processes and resources in the system, respectively. The deadlock detection latency is thus minimized end is constant regardless of m and n. Release handling tskes longer than conventional detection schemes. However, in a multiprocessor system environment, operating system can handle the release on the fly running on a seperate processor, thus not interfering with user process execution. To some class of applications, a predictable and zero-latency deadlock detection scheme could be very useful.