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 Resource Allocation Group
Part of ProcSimity visualization tool.

Resource Allocation Group

We are the Resource Allocation group at the University of Oregon Department of Computer and Information Science.

Keywords: Parallel and distributed computing, resource management, allocation, mapping, task assignment, scheduling, distributed shared memory.

Contents

  1. Research Description
  2. Project Archives (Technical Reports and Software)
  3. Group Members
  4. Other Information Resources
  5. Upcoming Conferences

Research Description:

Resource management is a key area of research in the drive to fully realize the performance potential of parallel and distributed computing systems. The complexities involved in managing both hardware and software resources that number in the hundreds to thousands, under the demands of a diverse multi-user workload, presents a spectrum of challenging problems for operating systems designers.

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.

Current Group Members

Other Information Resources:

Our Metacomputing notes and links.

Ohter Resources:


kurtw@cs.uoregon.edu