Task scheduling algorithms for heterogeneous processors

Topcuoglu, Haluk Rahmi; Hariri, Salim; Wu, Min-You

doi:10.1109/hcw.1999.765092

Cited by 180 publications

(126 citation statements)

References 13 publications

Supporting

Mentioning

125

Contrasting

Unclassified

Order By: Relevance

“…The area of scheduling for heterogeneous architectures (i.e., within a single machine) is relatively new and has not been studied in detail yet. Some of the few papers on this topic [12,19] also agree on this lack of studies.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Predictive Runtime Code Scheduling for Heterogeneous Architectures

Jiménez

Vilanova²,

Gelado³

et al. 2009

High Performance Embedded Architectures and Compilers

View full text Add to dashboard Cite

Abstract. Heterogeneous architectures are currently widespread. With the advent of easy-to-program general purpose GPUs, virtually every recent desktop computer is a heterogeneous system. Combining the CPU and the GPU brings great amounts of processing power. However, such architectures are often used in a restricted way for domain-specific applications like scientific applications and games, and they tend to be used by a single application at a time. We envision future heterogeneous computing systems where all their heterogeneous resources are continuously utilized by different applications with versioned critical parts to be able to better adapt their behavior and improve execution time, power consumption, response time and other constraints at runtime. Under such a model, adaptive scheduling becomes a critical component. In this paper, we propose a novel predictive user-level scheduler based on past performance history for heterogeneous systems. We developed several scheduling policies and present the study of their impact on system performance. We demonstrate that such scheduler allows multiple applications to fully utilize all available processing resources in CPU/GPUlike systems and consistently achieve speedups ranging from 30% to 40% compared to just using the GPU in a single application mode.

show abstract

Section: Related Workmentioning

confidence: 99%

“…A few papers [12,19] do study the effect of scheduling for heterogeneous architectures. Similarly to the schedulers for distributed systems, programs are also represented as graphs with nodes corresponding to tasks in the program using information known a priori.…”

Section: Related Workmentioning

confidence: 99%

Predictive Runtime Code Scheduling for Heterogeneous Architectures

Jiménez

Vilanova²,

Gelado³

et al. 2009

High Performance Embedded Architectures and Compilers

View full text Add to dashboard Cite

show abstract

“…This algorithm is a commonly cited list-scheduling heuristic [14]. In [15], Dongarra et al discussed the fundamental properties of a good bi-objective scheduling algorithm, and proposed an approximation algorithm namely RHEFT, which is extended from HEFT algorithm [13] by considering the reliability and allows the user to subjectively choose a trade-off between high reliability and low overall execution time. Recursive Critical Path (RCP) algorithm utilizes the dynamic programming strategy and iteratively find critical path to minimize the overall execution time [16].…”

Section: Related Workmentioning

confidence: 99%

“…Streamline [7], a workflow scheduler for streaming data, takes dynamic nature of the Grid into account and takes application requirements, constraints, and resource availability into consideration for scheduling decisions. To achieve the reduced overall execution time, Heterogeneous Earliest Finish Time (HEFT) heuristic algorithm [13] was proposed to initially order all the tasks of a workflow in descending order of their upward rank values calculated as the sum of the execution time and the communication time of the tasks. This algorithm is a commonly cited list-scheduling heuristic [14].…”

Section: Related Workmentioning

confidence: 99%

Distributed Workflow Scheduling Under Throughput and Budget Constraints in Grid Environments

Cao

Zhu

Ding

2014

Job Scheduling Strategies for Parallel Processing

View full text Add to dashboard Cite

Abstract. Grids enable sharing, selection and aggregation of geographically distributed resources among various organizations. They are emerging as promising computing paradigms for resource and computeintensive scientific workflow applications modeled as Directed Acyclic Graph (DAG) with intricate inter-task dependencies. With the growing popularity of real-time applications, streaming workflows continuously produce large quantity of experimental or simulation datasets, which need to be processed in a timely manner subject to certain performance and resource constraints. However, the heterogeneity and dynamics of Grid resources complicate the scheduling of streaming applications. In addition, the commercialization of Grids as a future trend is calling for policies to take resource cost into account while striving to satisfy the users' Quality of Service (QoS) requirements. In this paper, streaming workflow applications are modeled as DAGs. We formulate scheduling problems with two different objectives in mind, namely either maximize the throughput under a budget/cost constraint or minimize the execution cost under a minimum throughput constraint. Two different algorithms named as Budget constrained RATE (B-RATE) and Budget constrained SWAP (B-SWAP) are developed and evaluated under the first objective; Another two algorithms named as Throughput constrained RATE (T P -RATE) and Throughput constrained SWAP (T P -SWAP) are evaluated under the second objective. Experimental results based on GridSim showed that our algorithms either achieved much lower cost with similar throughput, or higher throughput with similar cost compared with other comparable existing algorithms.

show abstract

“…Scheduling task graphs on heterogeneous platforms Several heuristics have been introduced to schedule (acyclic) task graphs on different-speed processors, see [31,32,43,38,12] among others. Unfortunately, all these heuristics assume no restriction on the communication resources, which renders them somewhat unrealistic to model real-life applications.…”

Section: Related Problemsmentioning

confidence: 99%

Assessing the Impact and Limits of Steady-State Scheduling for Mixed Task and Data Parallelism on Heterogeneous Platforms

Beaumont

Legrand

Marchal

et al.

Third International Symposium on Parallel and Distributed Computing/Third International Workshop on Algorithms, Models and Tool

View full text Add to dashboard Cite

In this paper, we consider steady-state scheduling techniques for mapping a collection of application graphs onto heterogeneous systems, such as clusters and grids. We advocate the use of steady-state scheduling to solve this difficult problem. While the most difficult instances are shown to be NP-complete, most situations of practical interest are amenable to a periodic solution which can be described in compact form (polynomial size) and is asymptotically optimal.Keywords: Mixed parallelism, Scheduling, steady-state, heterogeneous platforms Résumé Dans ce rapport, nous appliquons des techniques d'ordonnancement en régime permanent pour ordonnancer une suite de graphes d'application sur une plateforme hétérogène de type "grille de calcul". Nous prônons l'usage de l'ordonnancement en régime permanent pour résoudre ce problème difficile. Nous montrons que les instances les plus ardues de ce problème sont NP-complètes, alors que la plupart des situations présentant un intérêt pratique peuventêtre réso-lues par une solution périodique qui admet une description compacte (de taille polynomiale) et qui est asymptotiquement optimale.

show abstract

Task scheduling algorithms for heterogeneous processors

Cited by 180 publications

References 13 publications

Predictive Runtime Code Scheduling for Heterogeneous Architectures

Predictive Runtime Code Scheduling for Heterogeneous Architectures

Distributed Workflow Scheduling Under Throughput and Budget Constraints in Grid Environments

Assessing the Impact and Limits of Steady-State Scheduling for Mixed Task and Data Parallelism on Heterogeneous Platforms

Contact Info

Product

Resources

About