WHP-0005 Virtualization Management for Grids and SOA

Virtualization introduces an abstraction layer on top of resources, so that physical characteristics are hidden from the user. At present CPU, network, and storage virtualization are the most widespread forms. Academic and/ or commercial data processing centres already use or are at the verge of adopting those virtualization technologies - mostly to increase their overall system utilization and to offer highly available services. In the context of Grid Computing and SOA a broader field of application is possible. This paper discusses how Grid Computing and SOA can take advantage of resource virtualization by integrating it at different levels of the software stack. The study focuses on the impact in the area of resourcemanagement and scheduling by supporting virtualization. To this end, the meaning of jobs is extended towards a broader meaning. It is proposed to move in Grid Computing from simple job submission towards a (more challenging and complex) submission of virtual machines. Necessary changes, requirements and problems at the Grid Middleware and the LRMS level are further discussed. In the long run both, user and resource provider, benefit from the adoption and combination of these technologies. Users no longer need to modify their application software depending on the offered execution environment at a resource. They just once create a virtual machine and directly include the necessary execution environment and the application level software as well. Subsequently the virtual machine can be submitted to every capable resource without requiring adjustments. On the other hand the virtualization layer reliefs the resource providers from selecting a fixed and limited execution environment and therefore increases the flexibility and system utilization as well. Due to strong interlinks between Grids and SOA results in the field of Grid Computing can be naturally extended to SOA environments. Similarly, the deployment and hosting of a service-based application can be managed within a virtual environment. Typically applications in SOA follow more a transaction-oriented, long running scenario in comparison to the typical space-shared job execution in most Grids. Here, the advantages of managing virtual environments with dynamic adaptation of the resource allocation is a major advantage too.