next up previous
Next: Split-C Application Benchmarks Up: Active Messages Implementation Previous: Overheads

Summary and Comparison with MPL

   table236
Table 3: Performance Summary of SP AM and IBM MPL

MPL's user space round-trip latency is measured with a ping-pong benchmark which sends a one-word message back and forth between two processing nodes using mpc_bsend and mpc_recv. The asymptotic bandwidth of SP AM is comparable to MPL's asymptotic bandwidth. A half-power point comparison indicates that SP AM delivers better performance than MPL for short messages.



Chris Hawblitzel
Thu Sep 19 12:22:33 EDT 1996