Implementation of reductions in support of PDES on a network of workstations
PADS '98 Proceedings of the twelfth workshop on Parallel and distributed simulation
Hi-index | 0.00 |
Reduction operations are very useful in parallel anddistributed computing, with applications in barriersynchronization, distributed snapshots, terminationdetection, global virtual time computation, etc. In thecontext of parallel discrete-event simulations, we havepreviously introduced a class of adaptive synchronizationalgorithms based on fast reductions. Here, we explorethe implementation of fast reductions on a popular highperformance computing platform - a network ofworkstations. The specific platform is a set of PentiumPro PC's running the Linux operating system, inter-connectedby Myrinet - a Gbps network. The generalreduction model on which our synchronizationalgorithms are based is introduced first, followed by adescription of how this model can be implemented. Wediscuss several design trade-offs that must be made inorder to achieve the driving goal of high speedreductions and provide innovative algorithms to meet thecorrectness and performance requirements of thereduction model.