Processor allocation and fault tolerance in multiprocessors /
With the speed of electronic devices approaching that of light, the use of multiple processors interconnected by a network offers the promise of a quantum leap in computing power for solving very large computational problems. The development of such multiprocessor systems has been accelerated by th...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
1995.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=742145011&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | With the speed of electronic devices approaching that of light, the use of multiple processors interconnected by a network offers the promise of a quantum leap in computing power for solving very large computational problems. The development of such multiprocessor systems has been accelerated by the improved price/performance ratio, compared to that of mainframes. When the number of processors in a multiprocessor system increases, two issues, namely subsystem allocation and fault tolerance, become very important. Subsystem allocation is essential to a multiprocessor system that is intended to support a multiuser environment. Similar to memory allocation, the primary objectives of subsystem allocation are to maximize the processor utilization and to minimize processor fragmentation in order to accommodate all the incoming tasks. Since a large multiprocessor system is vulnerable to faults, fault tolerance is extremely important in order to achieve sustained high performance. Since hypercube and multistage interconnection network (MIN) are two of the most popular architectures used in multiprocessor systems, this dissertation addresses the following issues related to subsystem allocation and fault tolerance: (1) Design of a processor allocation strategy to reduce processor fragmentation and maximize system utilization, (2) Development of a partitioning technique for MIN'S, (3) Development of a fault-tolerant reconfiguration technique for hypercubes and a fault-tolerant subcube allocation scheme, and (4) Subsystem reliability analysis for hypercubes and MIN'S. A new strategy using the concept of processor weights is proposed for subcube allocation. The weight of a processor is defined as the number of allocated processors that are the immediate neighbors of the processor. In addition, a systematic methodology is designed for partitioning a class of MIN's in order to facilitate subsystem allocation. The novel fault-tolerant reconfiguration technique is developed by taking advantage of the distant-insensitive wormhole routing technique. Based on the reconfiguration technique, a fault-tolerant subcube allocation is also proposed. Finally, two combinatorial reliability models, probability fault model and random fault model are proposed to evaluate the subsystem reliability of hypercubes and MIN'S. |
|---|---|
| Item Description: | Vita. "Major Subject: Computer Science". |
| Physical Description: | xiv, 151 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilms Inc. |
| Bibliography: | Includes bibliographical references. |