Folding contiguous and noncontiguous allocation policies for two-dimensional mesh multicomputer

Abstract EN

Current research results suggest that noncontiguous space sharing processor allocation is superior to contiguous space sharing allocation in two-dimensional mesh multi computers.

However, high system utilization is not achieved by both approaches due to the problem of fragmentation.

Policies that can execute applications on fewer processors than they have requested are proposed.

The goal of application folding is to reduce fragmentation and to increase program efficiency.

Simulation experiments that measure system utilization, response times, and interconnection contention are performed for two-dimensional mesh multicomputers with wormhole routing.

The results show that load-dependent folding substantially improves the performance of both contiguous and noncontiguous allocations.

However, contiguous allocation benefits more from folding, and its performance can be superior to that of noncontiguous allocation.