Scalable S-to-P broadcasting on message-passing MPPs

In s-to-p broadcasting, s processors in a p processor machine contain a message to be broadcast. We describe a number of different broadcasting algorithms that handle all ranges of s. We show how the performance of each algorithm is influenced by the distribution of the s source processors and the relationship between the distribution and the dimension of the machine. Each algorithm has ideal distributions and distributions on which the performance degrades. To reduce the dependence of the performance on the distribution of sources we propose a repositioning approach. In this approach the initial distribution is turned into an ideal distribution of the target broadcasting algorithm. We report experimental results for the Intel Paragon and discuss scalability and performance