The wide-scale deployment of parallel applications that deliver reliable high performance in multiprogrammed environments will require several technical innovations that are the focus of this research project. Specifically, this research is based on a new high-performance virtual machine model. In addition to the programming abstractions that traditionally define a virtual machine, our virtual machine's definition will include performance abstractions and every implementation of the virtual machine must support these performance abstractions. Therefore, we are investigating and developing algorithms for resource management, thread scheduling, and fault tolerance, all of which are necessary in the construction of such a virtual machine. Our goal is that programmers will be able to write "resource-oblivious" codes that nevertheless use non-dedicated resources with guaranteed end-to-end performance.
Overview | People | Papers |
---|