Hardware for fast global operations on multicomputers

The results presented show that the addition of a secondary network with a wide-tree topology and one or more coordination processors to a message-passing-only multicomputer can speed up synchronization, broadcasting, global reduction, and other collective communication operations by two or three orders of magnitude. While the idea of a secondary network is in itself not new, modeling and simulation show that our system is faster, more cost effective, and more versatile than any others we have found. As as direct evidence of this latter point, our system is applicable to workstation cluster multicomputers as well as to "Big Iron" type multicomputers. Although our system provides a large performance gain for global operations, it increases overall system cost by only 2-3%.<
>