We describe an algorithm for barrier synchronization that requires only read and write to shared store. The algorithm is faster than the traditional locked counter approach for two processors and has an attractive log 2 N time scaling for larger N. The algorithm is free of hot spots and critical regions and requires a shared memory bandwidth which grows linearly with N, the number of participating processors. We verify the technique using both a real shared memory multiprocessor, for numbers of processors up to 30, and a shared memory multiprocessor simulator, for number of processors up to 256.

No abstract

No abstract

No abstract

This report has been reproduced directly from the best available copy.