Performance evaluation of scheduling precedence-constrained computations on message-passing systems

Abstract

Scheduling precedence graphs with communication times is the theoretical basis for achieving efficient parallelism in message-passing machines. The lack of global information on the tasks, due to communication, has lead to develop local scheduling heuristics such as the Earliest-Task-First. Using knowledge on computation, communication, and system topology, a class of global priority-based scheduling heuristics called Generalized List Scheduling is proposed. The task-level is evaluated by backward scheduling the computation over the multiprocessor by using the best local heuristic. This leads to realistic measurement of the task priority for use in forward GLS scheduling.Experimental evaluation of local and GLS heuristics is carried out using extensive random graph generation and altering over the communication, inherent parallelism, and system topology. Analysis shows that local heuristics rely on locally maximizing the processor efficiency and gives acceptable deviations only when the inherent parallelism is large enough to cover the effective communication. This leads the local heuristics to achieve bounded speedup.GLS scheduling is based on combining two strategies: 1) differentiate critical computation and communications from others by scheduling critical paths first, and 2) implement effective management of processor utilization in order to increase the speedup. GLS scheduling maintains acceptable relative deviation versus change in parallelism, communication, and multiprocessor topology. The time complexity of GLS heuristics is O(pn 2 ) , where p and n are the number of processors and that of the tasks, respectively.

Evolution-Based Scheduling of Computations and Communications on Distributed-Memory Multicomputers

“…The complexity of ETF [13], DLS [4], and GLS [14] is O(pn 2 ), where n and p are the number of tasks and the number of processors, respectively. The initialization of edge addition SE (step 1 of SE-Schedule) is dominated by O(pn 2 K), where K is the number of forward-backward iterations of GLS and O(pn 2 ) is the complexity of one GLS scheduling iteration.…”

“…Generalized list scheduling (GLS) [14,15] uses a decision function based on highest-level-earliest-task-first (HLETF) which augments the ETF discipline with a task-level priority function to improve its performance. GLS iteratively schedules the forward and backward computation graphs.…”

“…Though [14] GLS was originally proposed for twoiteration scheduling we use it here as an iterative scheduling approach with an arbitrary number of iterations to allow the exploration of a space of solutions. The main advantage of GLS is its ability to find a local solution with a 'probable refinement' of the solution finish time via the iterations.…”

“…GLS [14] is used here for searching local schedules in order to evaluate the average mobility. GLS generates a solution in each iteration.…”

Modeling and Optimisation of Precedence-Constrained Production Sequencing and Scheduling for Multiple Production Lines Using Genetic Algorithms

“…Mouhamed & Maasarani [17] proposed a class of global priority-based scheduling heuristics, called generalized list scheduling, to schedule precedence-constrained computations on message-passing systems.…”