The use of virtual machines for Escience has been advocated both within the enterprise to replace aging machines and as the underlying technology of cloud computing whereby scientific researchers can ?rent? servers on demand. However, both scenarios can lead to inadequate performance. Within the enterprise, with incorrect planning or under unexpected heavy or even moderate load, there might not be enough physical capacity for every virtual machine to achieve reasonable performance. In cloud-computing-based scenarios, the ?renters? are largely subject to the informal service promises of the cloud provider based on a granularity that can be too coarse or at the wrong level of abstraction. This project pursues a novel unified framework to ensure predictable Escience based on these two dominant emerging uses of virtualized resources. The foundation of the approach is to wrap an Escience application in a performance container framework and dynamically regulate the application?s performance through the application of formal feedback control theory. The application?s progress is monitored and ensured such that the job meets its performance goals (e.g., deadline) without requiring exclusive access to physical resources even in the presence of a wide class of unexpected disturbances. This project extends this foundation and early results in three important dimensions: creating support for non-specialists to use the framework; implementing these techniques in Eucalyptus, one of the major open-source cloud computing frameworks; and applying the techniques to ?Software-as-a-Service? (SaaS), in which applications in the cloud are regulated to provide predictable performance.
