W. Chantamas and J.W. Baker (USA)
Associative Computing, Joint Data and Control Parallelism, Model Simulation, Large Scale Computing
The Multiple Associative Computing (MASC) parallel model is a generalization model of an Associative Computing (ASC) parallel model designed to support multiple ASC data parallel threads by using control parallelism. The MASC model is designed to combine the advantages of both Single Instruction Stream Multiple Data Streams (SIMD) and Multiple Instruction Streams Multiple Data Streams (MIMD) models. Here is the first time that a complete description of MASC model has been implemented (in software) true to its original description. A cycle precision simulator is built to demonstrate the performance of MASC on various multithreaded algorithms. The simulator is a software prototype for the model with sufficient software details to allow it to be converted into a hardware prototype of the model. If a reasonable limit for the number of threads simultaneously supported is assumed, the resulting hardware design is not only easily to implement, but can easily support a huge number of processing units and is a excellent candidate architecture for supporting large scale (e.g., terascale and petascale) computing. Experimental results shows that, when processing large-scale instances using multiple workers, the algorithm executed by the MASC model using a static task assignment scheme provides strong scaling with constant time overhead.
Important Links:
Go Back
