Title :
Reconfigurable mesh-connected processor arrays using row-column bypassing and direct replacement
Author :
Tsuda, Nobuo ; Shimizu, Tatsuyuki
Author_Institution :
Comput. & Network Syst. Core, Kanazawa Inst. of Technol., Ishikawa, Japan
Abstract :
This paper proposes an advanced reconfiguration scheme using row-column bypassing and direct replacement for two-dimensional mesh-connected processing-node arrays that makes an array usable for massively parallel computing and stand-alone computing in an efficient divided manner. This scheme uses an array providing a switching circuit in every node for row-column bypassing and a simple bypass network with a tree structure allocated to the array by graph-node coloring with a minimum inter-node distance of three for direct replacement. It can reconfigure a subarray with a regular matrix of free nodes usable for parallel computing in the array while allowing a small delay in the mesh connections but maintaining a communication path from every busy node being used as stand-alone computing to the outside of the array. The direct replacement is used for substitution of busy nodes which are not covered by row-column bypassing with free nodes located in the rows or columns to be bypassed, helping to enlarge the size of the reconfigured subarray. The bypass allocation with a minimum distance of three enables distributed communications and simple routing in the array while attaining a large success probability of the direct replacement. The proposed scheme is advantageous for constructing fault-tolerant massively parallel systems by using personal computers or workstations as processing nodes and Ethernet devices for interconnections
Keywords :
fault tolerant computing; graph colouring; multiprocessor interconnection networks; network routing; parallel architectures; reconfigurable architectures; workstation clusters; 2D mesh-connected processing-node arrays; Ethernet devices; communication path; direct replacement; distributed communications; fault-tolerant massively parallel systems; graph-node coloring; interconnections; massively parallel computing; personal computers; reconfigurable mesh-connected processor arrays; regular matrix; routing; row-column bypassing; stand-alone computing; switching circuit; tree structure; workstations; Concurrent computing; Delay; Fault tolerant systems; Matrices; Microcomputers; Parallel processing; Routing; Switching circuits; Tree data structures; Workstations;
Conference_Titel :
Parallel Architectures, Algorithms and Networks, 2000. I-SPAN 2000. Proceedings. International Symposium on
Conference_Location :
Dallas, TX
Print_ISBN :
0-7695-0936-3
DOI :
10.1109/ISPAN.2000.900256