Scheduling precedence graphs in systems with interprocessor communication times
SIAM Journal on Computing
Using duplication for scheduling unitary tasks on m processors with unit communication delays
Theoretical Computer Science
Optimal Scheduling Algorithm for Distributed-Memory Machines
IEEE Transactions on Parallel and Distributed Systems
Worst-case analysis of scheduling heuristics of parallel systems
Parallel Computing
On Exploiting Task Duplication in Parallel Program Scheduling
IEEE Transactions on Parallel and Distributed Systems
Approximation of algorithms for scheduling trees with general communication delays
Parallel Computing - Special issue on task scheduling problems for parallel and distributed systems
On the Granularity and Clustering of Directed Acyclic Task Graphs
IEEE Transactions on Parallel and Distributed Systems
A task duplication scheme for resolving deadlocks in clustered DAGs
Parallel Computing
Advantage of overlapping clusters for minimizing conductance
LATIN'12 Proceedings of the 10th Latin American international conference on Theoretical Informatics
Hi-index | 0.00 |
This article is concerned with the problem of scheduling a parallel application depicted by a precedence graph in presence of large communication delays. The target architecture is constituted of a bounded number m of identical processors linked together by an interconnection network. Communication delays represent the time of data transfer between two tasks of the application allocated to different processors. Our objective is to find an allocation of tasks to the processors and an execution order on each machine such that the overall completion time is minimized. We consider the special case of unit execution time for all computation tasks and a uniform communication delay 驴. We present a new approach based on the reduction of the problem to the successive schedulings of "small graphs", roughly speaking graphs which can be scheduled in time at most 驴 + 1 on an unbounded number of processors. Allowing duplication, corresponding to the recomputation of some of the tasks, this technique allows us to derive an asymptotic O(ln 驴/ ln ln 驴)-approximation algorithm for general precedence graph structure.