A resource management technique for processing deadline-constrained multi-stage workflows

Norman Lim,Shikharesh Majumdar,Peter Ashwood-Smith

doi:10.1186/s13677-017-0091-2

Abstract

The use of cloud computing that provides resources on demand to various types of users, including enterprises as well as engineering and scientific institutions, is growing rapidly. An effective resource management middleware is necessary to harness the power of the underlying distributed hardware in a cloud. Two of the key operations provided by a resource manager are resource allocation (matchmaking) and scheduling. This paper concerns the problem of matchmaking and scheduling an open stream of multi-stage jobs (or workflows) with Service Level Agreements (SLAs) on a cloud or cluster. Multi-stage jobs require service from multiple system resources and are characterized by multiple phases of execution. This paper presents a resource allocation and scheduling technique called RM-DCWF: Resource Management Technique for Deadline-constrained Workflows that can efficiently matchmake and schedule an open stream of multi-stage jobs with SLAs, where each SLA is characterized by an earliest start time, an execution time, and a deadline. A rigorous simulation-based performance evaluation of RM-DCWF is conducted using synthetic workloads derived from real scientific workflows. In addition, the impact of various system and workload parameters on system performance is investigated. The results of this performance evaluation demonstrate the effectiveness of RM-DCWF as captured in a low number of jobs missing their deadlines.

Highlights

Over the past few years, distributed computing paradigms such as cluster computing and cloud computing have been generating a lot of interest among consumers and service providers as well as researchers and system builders
To the best of our knowledge, none of the existing work focuses on all aspects of the resource management problem that this paper focuses on: devising a resource allocation and scheduling algorithm for multistage jobs with service level agreement (SLA) on a system subjected to an open stream of job arrivals
To the best of our knowledge, none of the existing work focuses on all aspects of the resource management problem that this paper focuses on: matchmaking and scheduling an open stream of multi-stage jobs with SLAs, where each SLA is characterized by an earliest start time, an execution time and an end-to-end deadline, on a distributed computing environment, such as a set of resources acquired a priori from a public cloud

Summary

Introduction

Over the past few years, distributed computing paradigms such as cluster computing and cloud computing have been generating a lot of interest among consumers and service providers as well as researchers and system builders. Investigating and devising effective resource management techniques for clouds and clusters is necessary to harness the power of the underlying distributed hardware and to achieve the performance objectives of a system [3], which. Once a number of jobs are allocated to a specific resource, a scheduling algorithm determines the order in which jobs allocated to the resource should be executed for achieving the desired system objectives. Performing effective matchmaking and scheduling is difficult because the SLA of the jobs need to be satisfied, while considering system objectives, which can include minimizing the number of jobs that miss their deadlines as well as generating adequate revenue for the service provider

Methods

Findings

Conclusion