On cluster resource allocation for multiple parallel task graphs

Abstract

a b s t r a c tMany scientific applications can be structured as parallel task graphs (PTGs), that is, graphs of data-parallel tasks. Adding data parallelism to a task-parallel application provides opportunities for higher performance and scalability, but poses additional scheduling challenges. In this paper, we study the off-line scheduling of multiple PTGs on a single, homogeneous cluster. The objective is to optimize performance without compromising fairness among the PTGs. We consider the range of previously proposed scheduling algorithms applicable to this problem, from both the applied and the theoretical literature, and we propose minor improvements when possible. Our main contribution is an extensive evaluation of these algorithms in simulation, using both synthetic and real-world application configurations, using two different metrics for performance and one metric for fairness. We identify a handful of algorithms that provide good trade-offs when considering all these metrics. The best algorithm overall is one that structures the schedule as a sequence of phases of increasing duration based on a makespan guarantee produced by an approximation algorithm.

Fair Resource Sharing for Dynamic Scheduling of Workflows on Heterogeneous Systems

“…In [7], the algorithms for off-line scheduling of concurrent parallel task graphs on a single homogeneous cluster were evaluated extensively. The graphs, or workflows, that have been submitted by different users share a set of resources and are ready to start their execution at the same time.…”

A Framework for Multiple Parallel Task Graphs (PTG) Scheduler

“…An extensive evaluation of algorithm proposed in [12] with some enhancements is carried out by Casanova et al [13]. They have also included k-shelf algorithm [14] for performance comparison.…”

Scheduling Workflows in Multi-cluster Environments

“…The interconnection speed between nodes in the same cluster is an average of 1 GB/s and the interconnection speed between nodes of different clusters is an average of 100 MB/s. We modeled a set of workflows using a script that follows the approach of automatic DAG generators [2]. The workflow generation is defined by the following parameters: number of levels, minimum and maximum number of tasks per level.…”

“…Such applications can be organized as workflows in order to be scheduled, executed, and managed in a distributed infrastructure such as multi-clusters. In this sense, the task parallelism and data parallelism [2], [3] can be explored to improve the application's performance. The task parallelism is present at level of tasks of a workflow, and the data parallelism can be present in a single task of a workflow.…”