MIMD Implementation on a Heterogeneous Network of Processors


"Microprocessors speed and efficiency are increasing dramatically, but the assumption is that the trend can't continue at current rates forever", according to Andy and Brad. Their presentation focused on heterogeneous networks of parallel processors. Some of the key topics that they covered were SIMD, MIMD Shared and MIMD Distributed.

Each node executes the same instruction stream in SIMD. In MIMD shared, each node executes the same or different instruction stream. But, Andy and Brad seemed to think that MIMD distributed was better to use that the two mentioned. Why? Performance. They believe that the MIMD distributed works better that the MIMD Shared. MIMD distributed works by using a message passing interface (MPI) which is a standard or communication protocol and the SPMD (single-program multiple-data) model. Andy and Brad covered various programs for MPI's such as: Point to Point Communication, Collective Communication, and other Communication Groups. They also compared MPI vs. FORCE. They are similar in that they both use the SPMD model, both are portable and both use a well known language. They are different in that MIMD shared uses the FORCE model and MIMD distributed uses MPI model. MIMD shared has shared variables and MPI has only private variables. Andy and Brad covered the topic extensively and also went into great detail about the MPI model, since they both believe that MIMD distributed is the way to go. They also wrote two programs and discussed an MPI host file. All in all, their presentation was very informative and explained MIMD on a heterogeneous network of processors thoroughly.

See details from the presentation.