In large-scale distributed memory multiprocessors, communication latency scaleup is the bottleneck of the overall system performance. Router design is the major factor that affects system communication performance. On a medium level of communication workload, queuing delay due to congestion can drastically increase communication latency. In this research, we designed a router of a globally adaptive routing method. The routing method polls traffic states in the mesh interconnection network. Based on the polled traffic information, it performs congestion-reduction routing and improves communication latency on a medium level of communication workload. The method partitions a mesh into a hierarchical structure to facilitate traffic state polling and routing at various levels of routing regions. The method adopts two detailed routing methods : look-around routing and parallel approximate maze routing, in order to make congestion-avoidance routing decisions at corresponding levels of routing regions. In this research, we proposed a system design of its router design. In addition, we added a multicast routing function and a fault-tolerance capability into the system design. We utilized system simulation of its hardware architectonic to compare the congestion-reduction capabilities of the router design and other commonly used router designs such as the minimal router, and thin overall communication performance.
Identifer | oai:union.ndltd.org:NSYSU/oai:NSYSU:etd-0909104-163702 |
Date | 09 September 2004 |
Creators | Wu, Chao-Yi |
Contributors | Chih-Chien Chen, Tsung Lee, Chia-Hsiung Kao |
Publisher | NSYSU |
Source Sets | NSYSU Electronic Thesis and Dissertation Archive |
Language | Cholon |
Detected Language | English |
Type | text |
Format | application/pdf |
Source | http://etd.lib.nsysu.edu.tw/ETD-db/ETD-search/view_etd?URN=etd-0909104-163702 |
Rights | unrestricted, Copyright information available at source archive |
Page generated in 0.0015 seconds