Issue No. 11 - November (1994 vol. 5)
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/71.329673
<p>Several techniques have been proposed to allow parallel access to a shared memorylocation by combining requests. They have one or more of the following attributes:requirements for a priori knowledge of the request to combine, restrictions on the routingof messages in the network, or the use of sophisticated interconnection network nodes.We present a new method of combining requests that does not have the aboverequirements. We obtain this new method for request combining by developing aclassification scheme for the existing methods of request combining. This classificationscheme is facilitated by separating the request combining process into a two partoperation: determining the combining set, which is the set of requests that participate ina combined access; and distributing the results of the combined access to the membersof the combining set. The classification of combining strategies is based upon whichsystem component, processor elements, or interconnection network performs each ofthese tasks. Our approach, which uses the interconnection network to establish thecombining set and the processor elements to distribute the results, lies in an unexploredarea of the design space. We also present simulation results to assess the benefits of theproposed approach.</p>
Index Termsmultiprocessor interconnection networks; parallel architectures; shared memory systems;virtual machines; message passing; multiprocessors; arbitrary interconnection networks;parallel access; shared memory location; message routing; classification scheme;combining strategies; combining set; processor elements; design space; simulationresults; hot spots
G. Sohi and A. Lebeck, "Request Combining in Multiprocessors with Arbitrary Interconnection Networks," in IEEE Transactions on Parallel & Distributed Systems, vol. 5, no. , pp. 1140-1155, 1994.