Improving parallel executions by increasing task granularity in task-based runtime systems using acyclic DAG clustering

Abstract

The task-based approach is a parallelization paradigm in which an algorithm is transformed into a direct acyclic graph of tasks: the vertices are computational elements extracted from the original algorithm and the edges are dependencies between those. During the execution, the management of the dependencies adds an overhead that can become significant when the computational cost of the tasks is low. A possibility to reduce the makespan is to aggregate the tasks to make them heavier, while having fewer of them, with the objective of mitigating the importance of the overhead. In this paper, we study an existing clustering/partitioning strategy to speed up the parallel execution of a task-based application. We provide two additional heuristics to this algorithm and perform an in-depth study on a large graph set. In addition, we propose a new model to estimate the execution duration and use it to choose the proper granularity. We show that this strategy allows speeding up a real numerical application by a factor of 7 on a multi-core system.

Computing the Heaviest Conflict-free Sub-DAG in DAG-based DLTs

“…The order in which the conflicts are resolved is either in the decreasing (greedy max) or increasing (greedy min) order based on their cumulative weight. The source code for the evaluation and the generation of the figures is available online [7].…”

GraphOpt: Constrained-Optimization-Based Parallelization of Irregular Graphs

“…Several algorithms have been proposed for scheduling DAGs [43], [44], [45], which use either list-based or clustering-based scheduling heuristics, while GRAPHOPT takes a different approach of modeling the core routine of the tool as a constrained-optimization problem, allowing the use of open-source solvers. The constrained optimizationbased approach is explored in [46], [47].…”

Parallel kinetic scheme for transport equations in complex toroidal geometry

“…We have tested several implementations. For more details on the implementation, we refer for instance to [2], where the algorithm is parallelized with the StarPU runtime, or to [5] where we use a specialized DAG (Direct Acyclic Graph) clustering algorithm. The DAG algorithm relies on an OpenMP implementation.…”

“…The resulting scheme has also nice parallelization possibilities. In a poloidal plane, the blocktriangular linear systems resulting from the DG scheme that are well solved by an optimized task-based implementation [2,5]. In the toroidal direction, the transport equations are solved by a simple shift operator.…”