Graph-based subcube allocation and task migration schemes for hypercube systems

Abstract

For multi-user multi-tasking hypercube systems, there have been two fundamental problems in the processing of the incoming tasks: subcube allocation and task migration. In this thesis, we propose a new top-down subcube allocation scheme, called SIG (Subcube Information Graph) strategy, and also present a task migration scheme to solve the fragmentation problem in a hypercube. The SIG strategy is based on a heuristic and a undirected graph, called FC(Free Cube)-graph, whose vertices represent the available subcubes, and edges response time and internal/external fragmentation, and to generate the higher dimension subcube by the cycle detection in the FC-graph. When a subcube is released, a heuristic algorithm is also used to maintain the available subcubes as large as possible and to reduce the response time. To effectively utilize non-cubic topology on hypercubes, the SIG strategy is extended to support non-cubic allocation. Using the connectivity information among free subcubes in the FC-graph, the SIG strategy utilizes smaller subcubes which are physically adjacent to each other for non-cubic allocation.