Development of new congestion control algorithms for broadcast based multiprocessor architectures with multiple input queues

In this thesis, a congestion control algorithm to improve 64-node, 2-Dimensional (2-D) Simultaneous Optical Multiprocessor Exchange Bus (SOME-Bus) performance is proposed. The congestion control algorithm is tested via simulation under several synthetic traffic patterns using Optimized Network Engineering Tool (OPNET) Modeler. Performance measures such as average processor utilization, average channel waiting time, average input waiting time and average network response time have been collected before and after applying the algorithm. For Client-Server traffic model, the proposed algorithm is able to increase the average processor utilization, between 1.59% and 12.57% with 4 threads, and between 1.21% and 13.91% for 8 threads; decrease the average channel waiting time between 3.47% and 16.17% with 4 threads, and between 6.18% and 25.02% for 8 threads; the average input waiting time between 2.41% and 19.29% with 4 threads, and between 4.38% and 27.85% for 8 threads; the average network response time between 2.85% and 21.07% with 4 threads, and between 4.99% and 30.27% for 8 threads. For Asynchronous Message Passing traffic model, the proposed algorithm is able to increase the average processor utilization between 1.86% and 13.75% with 4 threads, and between 1.04% and 13.91% for 8 threads; decrease the average channel waiting time between 4.26% and 18.07% with 4 threads, and between 10.55% and 29.53% for 8 threads; the average input waiting time between 4.63% and 22.06% with 4 threads, and between 5.67% and 30.22% for 8 threads; the average network response time between 17.53% and 44.71% with 4 threads, and between 31.16% and 52.09% for 8 threads.