Investigation of the effect of message combining mechanism on the 2D some-bus architecture

To minimize the effect of hot-spots and latencies in broadcast-based architectures, messages are combined in the output channel queue to form a new message containing the payloads of all original messages. In this thesis, the performance of the message combining method is investigated on a 64-node 2D SOME-Bus architecture which is a low-latency and high-bandwidth broadcast-based fiber-optic interconnection network employing the message passing protocol. The performance of the method is evaluated by OPNET Modeler with various synthetic traffic patterns including hot-region, uniform, perfect-shuffle and bit-reverse. Performance measures such as average processor utilization, average channel waiting time and average network response time have been collected before and after applying the combining method in the output channel queue of a node for varying values of the ratio of the mean channel service time to mean thread run time (T/R) and under different traffic patterns. The results show that average network response times decrease by 8.81% to 20.62%, average channel waiting times decrease by 15.39% to 28.02% and average processor utilization increase by 5.12% to 12.71% after applying the combining method.