Measuring the performance of parallel message-based process architectures

Message-based process architectures are widely regarded as an effective method for structuring parallel protocol processing on shared memory multi-processor platforms. A message-based process architecture binds one or more processing elements with data messages and control messages received from applications and network interfaces. In this architecture, parallelism is achieved by simultaneously escorting multiple messages on separate processing elements through a stack of protocol tasks. The paper reports performance results from empirical comparisons of a connection-oriented TCP/IP protocol stack implemented using two different parallel message-based process architectures. These performance experiments measure the throughput, context switching, and synchronization exhibited by the two process architectures on a shared memory multi-processor platform. The experimental results demonstrate the extent to which the selection of a parallel process architecture affects protocol stack performance.