Previous [ 1] [ 2] [ 3] [ 4] [ 5] [ 6] [ 7] [ 8] [ 9] [ 10] [ 11] [ 12] [ 13] [ 14] [ 15] [ 16] [ 17] [ 18] [ 19] [ 20] [ 21] [ 22] [ 23] [ 24] [ 25]

@

Journal of Information Science and Engineering, Vol. 27 No. 2, pp. 527-544 (March 2011)

Injection Level Flow Control for Network-on-Chip (NoC)*

MINGHUA TANG1 AND XIAOLA LIN2,3,+
1School of Computer Science and Technology
GuangDong University of Finance
Guangzhou, 510521 P.R. China
2School of Information Science and Technology
3Key Laboratory of Digital Life
Ministry of Education
Sun Yat-sen University
Guangzhou, 510006 P.R. China

It is observed that NoC may saturate at certain rate of injecting packets, causing significant network performance degradation. The source node therefore should be notified about the network traffic condition to control its rate of injecting packets into the network accordingly. In this paper, we present a novel flow control strategy suitable for any routing algorithm by exploring the congestion status from a new viewpoint of share pattern. The proposed strategy is based on the concept of injection level, aiming at regulating the sending packet rates using several injection levels according to network status. A simple method is presented to obtain the network status. Implementing the strategy need not change the router design. Performance evaluation has been conducted based on flitaccurate and open source SystemC simulator. The results show that, using the proposed method, the NoC could run smoothly to avoid saturation problem. The average packet delay is dropped to the ideal level without hurting the throughput. A balance has been achieved between packet latency and network throughput. The energy consumption is also reduced since the proposed strategy significantly decreases the network payload.

Keywords: network-on-chip, OE routing algorithm, selection strategy, adaptive routing, flow control

Full Text () Retrieve PDF document (201103_09.pdf)

Received May 14, 2009; revised September 1, 2009 & March 1 & June 1, 2010; accepted June 4, 2010.
Communicated by Ren-Hung Hwang.
* The paper has been partially presented in the 8th IEEE/ACIS International Conference on Computer and Information Science, June 1-3, 2009, Shanghai, China and sponsored by IEEE Computer Society, Association of Computer and Information Science (ACIS) and Shanghai University.
+ Corresponding author.