Assume that a processor array can be dynamically partitioned into several subarrays each of which can be concurrently allocated to one of the incoming independent programs and, after completion of that program, reclaimed by the system for future use. Due to various processor requirements of programs, fragmentation of the array may occur and eventually decrease both array utilization and throughput. We describe an allocation method that, combined with a static optimizing mapping, reduces array fragmentation and improves system's throughput.