Date: Tue, 26 Nov 1996 00:57:06 GMT Server: Apache/1.1.1 Content-type: text/html Content-length: 5783 Last-modified: Tue, 19 Nov 1996 00:51:02 GMT
Keywords: Parallel and distributed computing, resource management, allocation, mapping, task assignment, scheduling, distributed shared memory.
Our research group focuses on the management of processors and processes in systems ranging from message-passing multicomputers to loosely-coupled workstation-based distributed systems. We have developed algorithms and software tools for allocation, mapping, placement, scheduling and migration, with extensions to support fault tolerance, heterogeneous environments, and real time constraints. We have also begun to explore issues related to parallel I/O allocation and scheduling. Our approach involves a blend of theoretical, experimental, and systems development work.
Processor allocation research: this work involves the design of processor allocation algorithms for message-passing machines based on the mesh and k-ary n-cube network topologies. Processor allocation involves selection of a subset of processors for assignment to each incoming job request with the goal of maximizing system throughput. A simulation and visualization tool called ProcSimity has been developed to support experimentation and performance analysis with a wide range of allocation algorithms on a spectrum of machine architectures. This work also involves empirical experimentation with state of the art machines through collaboration with Bill Nitzberg at NASA Ames NAS.
The OREGAMI project: involves the development of algorithms and abstractions for the mapping of parallel algorithms to message-passing machines when both the computation and the interconnection network are regular in structure. In collaboration with Sanjay Rajopadhye of IRISA , France, we have developed a formalism for describing both the computation and the target architecture which aids in the development of efficient and effective mapping functions. This approach exploits regularity in both the spatial and temporal communication patterns exhibited by many parallel applications.
Ohter Resources: