Analysis of communication data: compression network

The paper reports the effect of compressed message delivery (a compression network) on a parallel processing system such as workstation clusters. Compression network is simulated on a massively parallel processor, the Thinking Machines CM-5, using the parallel language Split-C. To examine the benefit of data compression to the system, profiles, including message pattern, number of messages, and message size, are measured by simulation. The results indicate that compression increases the network bandwidth. A theoretical compression ratio of about a half was estimated for almost all messages. Network bandwidth can be efficiently doubled using data compression.<
>