Keywords: parallel architecture, virtualisation, dynamic, heterogeneous, webcom.

The execution of many problems is constrained to a limited number of resources for the sake of efficiency. For example, in an high performance computing environment, a mismatch between the resources and the specific specific demands of an application can be costly and often results in performance degradation and resource underutilisation[1,2]. This paper presents WebComV, a framework to support the virtualisation of parallel resources in an heterogeneous, dynamic environment. WebComV, which is based on the existing WebCom [3] framework, acts as an idealised virtual machine, dynamically optimised for the execution of particular algorithms. The framework configures and maintains a static virtual view of available resources to the application irrespective of any dynamic behaviour of the underlying physical resources. The virtual view is specified by the algorithm writer and represents the ideal execution environment for the application. Related work, such as the provision of virtual workspaces [4] for the grid, illustrates that this research is particularly pertinent in a grid computing [5] environment.

This paper identifies and outlines three application categories, each with distinct virtualisation requirements, which will be supported within the WebComV framework. The WebComV architecture and its operation under various conditions of the underlying physical resources is described, illustrating how it builds on and exploits features within the existing WebCom framework.

Potential uses of the WebComV framework are identified in each of the following areas:

• Provision of a virtual parallel architecture in a 'cluster-of-clusters' environment, where cluster resource management (CRM) software (e.g., Torque[6]) interacts with WebComV allowing for seamless submission to multiple clusters managed by diverse CRM applications.
• Support and enhancement of applications which benefit from dynamic load balancing techniques[7]. Such applications have the potential to expand and shrink dynamically to accommodate addition or removal of physical resources, which can be managed and virtualised by WebComV.
• Provision of 'virtual clusters' [1] by aggregating or partitioning available physical heterogeneous or homogeneous resources to support parallel execution with a parallel programming model such as the message passing interface.

1

I.T. Foster, T. Freeman, K. Keahey, D. Scheftner, B. Sotomayer, X. Zhang, "Virtual Clusters for Grid Communities", CCGRID, Pages 513-520, 2006.

2

M.F. Mergen, V. Uhlig, O. Krieger, J. Xenidis, "Virtualization for high-performance computing", Operating Systems Review, Volume 40, Pages 8-11, 2006. doi:10.1145/1131322.1131328

3

J.P. Morrison, J.J. Kennedy, D.A. Power, "WebCom: A Web Based Volunteer Computer", The Journal of Supercomputing, Volume 18, pages 47-61, 2001. doi:10.1023/A:1008163024500

4

K. Keahey, I.T. Foster, T. Freeman, X. Zhang, D. Galron, "Virtual Workspaces in the Grid", Euro-Par 2005, pages 421-431, 2005.

5

I. Foster, C. Kesselman, "The Grid: Blueprint for a New Computing Infrastructure", Morgan Kaufmann Publ. inc., USA.

6

"TORQUE Resource Manager", www.supercluster.org/torque.

7

R. Cariño, I. Banicescu, "A Load Balancing Tool for Distributed Parallel Loops", Journal of Cluster Computing, Volume 8, Number 4, Pages 313-321, 2005. doi:10.1007/s10586-005-4098-y

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £95 +P&P)