|
09/14/2006, 9:30 AM - 11:00 AM
Speakers:
Kate Keahey, Computer Scientist, Argonne National Lab.
Tim Freeman, Developer, Researcher, University of Chicago.
One of the greatest challenges in Grid computing today is that while Grids provide access to many heterogeneous resources, a user's application will typically run only in a specific, customized software environment. Variations in operating systems, middleware versions, library environments, and filesystem layouts all pose barriers to application portability. Applications that work on a developer's desktop may function "out of the box" on only a small fraction of the total number of compute resources potentially available to the scientist. Another issue is lack of support for performance isolation: activities associated with one user or virtual organization (VO) can influence the performance seen by other processes executing on the same platform in an uncontrolled way. This inability to enforce QoS and provide execution guarantees prevents Grids from being useful in a range of scenarios.
To address those challenges we introduced the abstraction of a Virtual Workspace (VW) describing a virtual execution environment that can be made dynamically available to authorized Grid clients by using well-defined protocols. Virtual workspaces address two aspects of Grid environment: (1) providing a configurable execution environment and (2) associating it with a resource allocation -- described in terms of memory size or the number of processors -- that can be managed during the lifetime of a workspace. While workspaces can be implemented in a variety of ways, a particularly promising one is based on the virtual machine (VM) technology. VMs allow a client to create a custom execution environment configured with a required operating system, software stack and access policies and then deploy it on any resource running a hypervisor. VM state may be serialized into a VM image, allowing the client to pause or shut down VM operation, and resume it at a different time and location, thus enabling migration. Finally, VMs and the associated tools also offer excellent enforcement of resource usage. Modern hypervisors make all of these advantages available with little overhead to the application.
In this talk, we will describe the design, interfaces and implementation of the workspace service enabling the secure deployment and management of workspaces implemented as Xen virtual machines. We will discuss security aspects of workspace deployment and describe how deployed workspaces can be discovered and managed. Further, we will describe the generalization of workspaces to virtual clusters and the strategies for their mapping to site resources. Finally, we will describe the application of virtual workspaces to scientific as well as educational use cases. In particular, we will describe the use of workspaces in Open Science Grid's Edge Services project as well as their potential to provide a flexible and convenient vehicle to implement scientific gateways for application deployment.
|
