(SM)2-II
Authors:
Open Access
- 1 June 1985
- journal article
- Published by Association for Computing Machinery (ACM) in ACM SIGARCH Computer Architecture News
- Vol. 13 (3) , 100-107
Abstract
In order to treat scientific calculation with sparse matrices effectively, we propose a dedicated parallel machine called "the Sparse Matrix Solving Machine" version II, or (SM)2-II for short. A computation model named NC (NODE & CONNECTING-LINE) model and a new data-interchange method named RSM (Receiver Selectable Multicast) are introduced for (SM)2.1L In RSM, all processing units (PUs) share a global logical address. A PU can recognize whether or not it needs certain data from the address which the data is written into, and only data which is needed is actually written into memory. With this method a PU reads the data without conflict, and the requirement for memory is minimal. In this paper, congestion on the system bus is analyzed. Through this analysis the effectiveness of RSM will be justified.Keywords
This publication has 5 references indexed in Scilit:
- OCCAMACM SIGPLAN Notices, 1983
- Performance measurements on HEP - a pipelined MIMD computer1983
- (SM)2-Sparse Matrix Solving Machine1983
- Communicating sequential processesCommunications of the ACM, 1978
- The CRAY-1 computer systemCommunications of the ACM, 1978