Ordered Partial Task Compaction on Mesh Connected Computers

Task compaction has been examined as a means of reducing fragmentation in par titionable machines based on multi stage hypercube mesh and linear array inter connection networks Ordered partial task compaction involves moving a subset of executing tasks without permuting their relative order to accommodate a request for a large submesh that would otherwise be blocked from entering the system In this paper we develop the algorithms needed to nd good allocation sites and to perform one dimensional ordered partial compactions simply on mesh of processor and re con gurable mesh architectures We nd that signi cant performance gains can be obtained in heavily loaded systems even when link delays are large