Alternative Architectures

Before examining the architecture of the Grand Interconnect in detail, it is worth examining some of the commercial alternatives including FDDI and Ethernet. Both of these architectures are generally available and widely used. Both work very well for general purpose peer-to-peer processor links. They can be successfully used in data acquisition applications with suitably low performance requirements, and modest time coherency requirements between nodes. In general, they have good throughput, poor latency, and always require a relatively high performance processor to handle network protocol. Latency, the time from when one node attempts to initiate a transfer of information until the target accepts the transfer of the first byte, is a direct consequence of the peer-to-peer architecture of these networks. Furthermore, while "average latencies'' of such networks may be quite acceptable for some cases, worst case latencies can be substantial---even in relatively controlled environments.

In many high performance data acquisition applications, these latencies may not be acceptable and may result in some level of difficulty in correlating data from different nodes. The Grand Interconnect was developed to overcome these limitations as well as provide a lower cost solution for high performance data acquisition.