Low Complexity Performance Effective Task Scheduling Algorithm for Heterogeneous Computing Environments

Abstract

Abstract:A heterogeneous computing environment is a suite of heterogeneous processors interconnected by high-speed networks, thereby promising high speed processing of computationally intensive applications with diverse computing needs. Scheduling of an application modeled by Directed Acyclic Graph (DAG) is a key issue when aiming at high performance in this kind of environment. The problem is generally addressed in terms of task scheduling, where tasks are the schedulable units of a program. The task scheduling problems have been shown to be NP-complete in general as well as several restricted cases. In this study we present a simple scheduling algorithm based on list scheduling, namely, low complexity Performance Effective Task Scheduling (PETS) algorithm for heterogeneous computing systems with complexity O (e) (p+ log v), which provides effective results for applications represented by DAGs. The analysis and experiments based on both randomly generated graphs and graphs of some real applications show that the PETS algorithm substantially outperforms the existing scheduling algorithms such as Heterogeneous Earliest Finish Time (HEFT), Critical-Path-On a Processor (CPOP) and Levelized Min Time (LMT), in terms of schedule length ratio, speedup, efficiency, running time and frequency of best results.Key words: DAG, task graph, task scheduling, heterogeneous computing system, schedule length, speedup, efficiency INTRODUTIONA growing emphasis on concurrent processing of jobs has lead to an increased acceptance of heterogeneous computing environments and the availability of a network of processors makes a costeffective utilization of underlying parallelism for applications like weather modeling, image processing, real-time and distributed database systems. A wellknown strategy behind efficient execution of a huge application on a heterogeneous computing environment is to partition it into multiple independent tasks and schedule such tasks over a set of available processors. A task-partitioning algorithm takes care of efficiently dividing an application into tasks of appropriate grain size and an abstract model of such a partitioned application can be represented by a Directed Acyclic Graph (DAG). Each task of a DAG corresponds to a sequence of operations and a directed edge represents the precedence constraints between the tasks. Each task can be executed on a processor and the directed edge shows transfer of relevant data from one processor to another. Task scheduling can be performed at compile-time or at run-time. When the characteristics of an application, which includes execution times of tasks on different processors, the data size of the communication between tasks and the task dependencies, are known a priori, it is represented with a static model. The objective of task scheduling is to map the tasks on the processors and order their execution so that task precedence requirements are satisfied and a minimum overall completion time is obtained. The problem of scheduling of tasks with required precedence rela...

Task Scheduling Optimization in Heterogeneous Distributed Systems

“…Finally, a processor selection phase where the selected task is assigned to a processor in the set of processors that minimizes its finish execution time using the insertion-based scheduling policy [6].The algorithm has an O(np) time complexity for n nodes and p processors.…”

“…Average Computation Cost (ACC) and Maximum Data Arrival Cost (MDAC) compute this ECT. Next Equations (6,7,8) explains ACC, MDAC and ECT respectively.…”

“…Denotes the Earliest Finish Time of a task on a processor (EFT( , )) [6] and is defined in Equation (2).…”

“…Scheduling is done at run-time. In this paper, we focus on static scheduling [5,6]. Static scheduling is classified into list-based, clustering and duplication based.…”

Performance Enhancement of Scheduling Algorithm in Heterogeneous Distributed Computing Systems

“…In prioritization phase, the algorithms assign a certain priority that is computed, to node in the DAG. In machine assignment phase, each task depending on its priority is assigned to machine that minimizes the cost function [5][6][7][8][9]. Examples of list-based algorithms are Heterogeneous Earliest Finish Time (HEFT) and Critical Path on Processor (CPOP) [10].…”