| 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] |
¡@
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.
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.