PERFORMANCE OF HYPERCUBE ARCHITECTURE WITH SHARED-MEMORY

The performance of hypercube architectures depends on the topology of the problem being solved. Specifically, they are unsuitable for knowle dge-based applications where global data access is very often required . In this work. a new architecture is proposed that can be used in suc h applications. The proposed architecture uses both the shared memory and the distributed memories. In this architecture, shared memory is c onnected to the basic hypercubes using multistage interconnection netw orks (MINs). In this analysis, the Omega network is considered. When g lobal memory and distributed memories are used, multiple copies of sha red data deteriorate the system performance because of data invalidati on traffic in the network, and the coherency implementation overhead. Maintaining single copies of the shared data can eliminate the invalid ation traffic and the coherency related activities. However, the singl e copies are to be dynamically moved using a data movement policy to s erve the memory references. Analytical modelling of data movement poli cies is developed using Markov chains. The performance measures of the proposed architecture are compared with that of hypercube architectur e. The analysis includes measures for the proposed architecture with d ifferent basic hypercube sizes and compares them. In all cases, these measures demonstrate that the architecture is better than the traditio nal hypercube architecture for any given number of processing elements .