B.Robic, J.Silc.
Efficient Computation on Processor Arrays in a Multi-user Environment .
Proc. 22nd Euromicro Conference Beyond 2000: Hardware/Software Design Strategies, Prague, Czech Republic, September 2-5 1996.

We address the problem of performance improvement of a processor array in a multi-user environment. We assume that the array can be dynamically partitioned into subarrays that can be allocated to independent programs. When a program completes, it releases its subarray for future use. Since programs require subarrays of various sizes, an allocation method should be used that reduces array fragmentation and, as a result, improves processor utilization. If, before the allocation, the processor requirements of programs are minimized, then more programs can be placed on the array concurrently. This results in lower turn-around times for incoming programs. In this paper we present program compression, a compile-time method that minimizes processor requirements of programs, and subarray allocation, an allocation method that reduces array fragmentation. Experimental results show that efficiency of the computation on processor arrays can be considerably improved if compiler and operating system make use of program compression and subarray allocation, respectively.