Key Messaging on SOME-Bus clusters
Parallel Computing
| Hi-index | 0.00 |
Networks of Workstations have been mostly designedusing switch-based architectures and programming basedon message passing. This paper describes a network ofworkstations based on the Simultaneous OpticalMultiprocessor Exchange Bus (SOME-Bus) which is alow-latency, high-bandwidth interconnection network thatdirectly links arbitrary pairs of processor nodes withoutcontention, and can efficiently interconnect several hundrednodes. Each node has a dedicated output channel and anarray of receivers, with one receiver dedicated to everyother node's output channel. The SOME-Bus eliminates theneed for global arbitration and provides bandwidth thatscales directly with the number of nodes in the system.Under the Distributed Shared Memory (DSM) paradigm,the SOME-bus allows strong integration of the transmitter,receiver and cache controller hardware to produce a highlyintegrated system-wide cache coherence mechanism. Thispaper examines switch-based networks that maintain highperformance under varying degrees of application locality,and compares them to the SOME-Bus, in terms of latencyand processor utilization.