Performance effects of synchronization in parallel processors
Roger D. Chamberlain, Mark A. Franklin · 2002
We investigate synchronization activities in application executing on distributed-memory MIMD architectures. Three applications are used to quantify the performance impact of synchronization as the number of processors is increased. We also investigate the performance improvement possible when synchronization is supported in hardware. The results show that significant performance improvement can be achieved. The hardware support should include barrier synchronization, operate-and-broadcast, and operations over subsets of processors.>