Management of Virtual Machines in the Grid

Since the late 1990s, we have witnessed an extraordinary development of Grid technologies. Nowadays, different Grids are being deployed within the context of a growing number of national and transnational research projects (e.g. EGEE, TeraGrid or OSG). These projects have achieved unseen levels of resource sharing, offering a dramatic increase in the number of processing and storage resources that can be delivered to applications.

However, a growing heterogeneity on the organizations that joins these projects hinders the development of large scale Grid infrastructures. Grid resources do not only differ in their hardware but also in their software configurations (operating systems, libraries, and applications). This heterogeneity increases the cost and length of the applica tion development cycle, as they have to be tested in a great variety of environments where the developers have limited configuration capabilities. Therefore, some of the users are only able to use a small fraction of the Grid. Other projects, like EGEE, limit heterogeneity somewhat by imposing a fixed configuration for Grid resources. Moreover, most of the Grid infrastructures do not allow administrators to isolate and partition the performance of the resources they devote to the Grid. In this way, the applications of a grid user can affect the execution of other grid or local users. This limits the quality of service and reliability of actual platforms, preventing a wide adoption of the Grid paradigm.

Virtual machines add a new abstraction layer that allows partitioning and isolating the physical hardware resources. They also offer a natural way to face a highly heterogeneous environment.In the last years, the processors' performance has increased enough to renew the interest in the virtual machines technology. These technologies include operating system partitioning (e.g. Solaris Containers), complete hardware emulation (e.g. VMWare), or para-virtualization (e.g. Xen). Virtual machines presents attractive benefits like server consolidation, virtual machines isolation, performance partitioning or legacy applications execution among others.

We are following two directions to explore alternatives for the management of virtual machines in Grid infrastructures:

• We propose a straightforward deployment of virtual machines using the GridWay metascheduler to overcome some of the problems found in current Grid infrastructures, like heterogeneity, performance partitioning or application isolation. This strategy does not require additional middleware to be installed and it is not bounded to a virtualization technology. However, this solution presents several drawbacks like a limited use of the potential benefits offered by the virtualization technology (e.g. server consolidation), or a limited application range (HTC and workflow computations). This alternative is already functional.

• If you are interested in dynamic deployment and re-allocation of virtual machines you can go to GridHypervisor project.