VIRTUAL MACHINE

Virtual Machine

Abstract

There are many scientific applications that have high performance computing demands. Such demands are traditionally supported by cluster- or Grid-based systems. Cloud computing, which has experienced a tremendous growth, emerged as an approach to provide on-demand access to computing resources. The cloud computing paradigm offers a number of advantages over other distributed platforms. Our experimental results show that lightweight virtualization imposes about 5% overhead and it substantially outperforms traditional heavy-weight virtualization such as VMware.

Table of Content

Introduction4

Background5

Features and capabilities5

Description of selected application6

Distributed VTMs6

The Job Scheduler7

The VTM Scheduler8

The VTM Factory8

The Job Monitor9

Evaluation10

Experimental Results10

Findings14

Conclusion14

References16

Virtual Machine

Introduction

High performance computing (HPC) applications expand several scientific fields such as meteorology, astronomy, chemistry, and bioinformatics among others. Examples of these applications include weather forecasting, materials science, quantum chemistry calculations, accelerator modeling, and astrophysical simulations. These kinds of applications have huge demands of computing resources, which are traditionally supported by cluster- or Grid-based systems. Cloud computing, which has experienced a tremendous growth recently, emerged as an approach to provide on-demand access to computing resources. The cloud computing paradigm offers a number of advantages over other distributed platforms. For example, the access to resources is flexible and cost-effective since it is not necessary to invest a large amount of money on computing infrastructure nor pay salaries for maintenance functions (Antypas, 2008).

In this paper, we present an application-level virtualization framework, called Distributed Virtual Task Machine (D-VTM), which can significantly increase performance while retaining the benefits of cloud computing in terms of ease of management, cost reductions, resource usage optimization, etc.

Background

Much research work has focused on evaluating the suitability of virtualized environments for executing HPC applications. Wang (2009) presented comprehensive studies of the performance of parallel applications on the Amazon EC2 cloud platform and a Eucalyptus cloud system, respectively, using a set of typical HPC applications. Related cloud performance studies also considered several specific HPC applications including climate modeling, bioinformatics, lattice optimization, and astronomy workflows. These results all indicate that cloud execution incurs a significant performance overhead, particularly high for communication-intensive applications.

Features and capabilities

The D-VTM framework is based on application-level virtualization. This kind of virtualization is built on top of the operating system services. Hence, the same operating system image is used for all virtual machine (VM) instances (Juan, 2009). Abstract resources explicitly represent system resources. In addition, the framework was also extended with a number of VM management functions, as shown in Fig. 1. The Job Scheduler is in charge of receiving job requests accompanied with the desired number of VM instances. The VTM Scheduler keeps track of resource usage and determines if the required resources are available. The Job Monitor also keeps track of resource usage for billing purposes. Further details of each of these modules are presented below.

Description of selected application Distributed VTMs

Importantly, a distributed VTM is a specialization of composite VTMs whereby two or more address space is involved. Hence, a distributed VTM includes various local and remote primitive VTMs. Finally, the VTM scheduler and VTM factory are defined on a per-address ...