Scheduling precedence graphs in systems with interprocessor communication times
SIAM Journal on Computing
A threshold scheduling strategy for Sisal on distributed memory machines
Journal of Parallel and Distributed Computing
A Scalable Scheduling Scheme for Functional Parallelism on Distributed Memory Multiprocessor Systems
IEEE Transactions on Parallel and Distributed Systems
Program repartitioning on varying communication cost parallel architectures
Journal of Parallel and Distributed Computing
Partitioning and Scheduling Parallel Programs for Multiprocessors
Partitioning and Scheduling Parallel Programs for Multiprocessors
IEEE Transactions on Parallel and Distributed Systems
A fast and scalable scheduling algorithm for distributed memory systems
SPDP '95 Proceedings of the 7th IEEE Symposium on Parallel and Distributeed Processing
Scalable Scheduling Algorithm For Distributed Memory Machines
SPDP '96 Proceedings of the 8th IEEE Symposium on Parallel and Distributed Processing (SPDP '96)
Task scheduling algorithms for distributed memory systems
Task scheduling algorithms for distributed memory systems
A framework for performance-based program partitioning
Progress in computer research
A framework for performance-based program partitioning
Progress in computer research
Hi-index | 0.00 |
One of the major limitations of compile time scheduling schemes is the inability to precisely determine the computation and communication costs prior to generating the schedule. The authors address the issue of sensitivity of a given scheduling algorithm to the variations in imprecisely known compile time costs. The variations in the compile time costs can affect in one of the two ways: (i) original schedule found by the algorithm using estimated compile time costs does not change (schedule is invariant), or (ii) original schedule found by the algorithm changes when the costs change. For the first scenario, they have derived the conditions under which the schedule found by our algorithm would be invariant. For those cases where the schedule length changes, they have also introduced a measure of sensitivity of the schedule or the scheduling algorithm, defined as the ratio of percentage change in schedule length to that of maximum allowable percentage change in a node computation cost or an edge communication cost. Through an experimental study they show that the proposed algorithm is extremely insensitive and can be used in practical scheduling situations, where the compile time costs are known imprecisely.