M. Thottethodi, A. R. Lebeck, and S. S. Mukherjee. SelfTuned Congestion Control for Multiprocessor Networks. In Proceedings of the Seventh IEEE Symposium on HighPerformance Computer Architecture, Jan. 2001. Home/Search Document Details and Download
Summary Related Articles Check |
This paper is cited in the following contexts:
Bandwidth Adaptive Snooping - Martin, Sorin, Hill, Wood (2002) (5 citations) (Correct)
....theory to adjust network usage to avoid performance degradation due to hot spots in memory access patterns. This work differs from bandwidth adaptive snooping in that it seeks to throttle requests from being issued, whereas BASH chooses between issuing a unicast or a broadcast. Thottethodi et al. [31] have developed a scheme for using global network information to throttle requests before they can congest the network. At the cost of an additional network sideband for communication of contention effects, this scheme can adapt more accurately than a scheme that relies solely on local ....
....complements bandwidth adaptive snooping in that it could be used to estimate network utilization, and future work may adapt these techniques to improve our detection of network congestion. Other research has also explored techniques for estimating interconnect traffic (refer to the related work in [31]) 7 Conclusions and Future Work We have developed a hybrid shared memory cache coherence protocol, and we have demonstrated the benefits provided by adaptivity. Moreover, the trend towards integration of the coherence protocol logic and the processor on the same die suggests a unified adaptive ....
[Article contains additional citation context not shown here]
M. Thottethodi, A. R. Lebeck, and S. S. Mukherjee. SelfTuned Congestion Control for Multiprocessor Networks. In Proceedings of the Seventh IEEE Symposium on HighPerformance Computer Architecture, Jan. 2001.
Self-Tuned Congestion Control for Multiprocessor Networks - Thottethodi, Lebeck.. (2000) (2 citations) Self-citation (Thottethodi Lebeck Mukherjee) (Correct)
....for steady loads with uniform random communication pattern. We also present results for a bursty load with the various communication patterns. Results for steady loads with bit reversal, perfect shuffle and butterfly communication patterns are presented in greater detail in a technical report [28]. We use synthetic workload, instead of full blown multiprocessor workloads, for three reasons. First, our simulation environment cannot handle full blown multiprocessor workloads. Second, our packet generation frequency corresponds to realistic miss rates in databases and scientific ....
....buffers) 12 bits are enough to count all buffers in the network. Our configuration needs 13 bits to represent the maximum possible aggregate throughput g NodeCount MaxT raffic = 32 256 1 = 8192 flits) Thus, we need a total of 25 bits for the sideband signals. However, in our technical report [28], we show that we can send these 25 bits using 9 bit sideband channels with very little performance degradation. For comparison, we also simulate the At Least One (ALO) 2] congestion control scheme. ALO estimates global network congestion locally at each node. If at least one virtual channel is ....
[Article contains additional citation context not shown here]
M. Thottethodi, A. R. Lebeck, and S. S. Mukherjee. SelfTuned Congestion Control for Multiprocessor Networks. Technical Report CS-2000-15, Duke University, November 2000.
Using Speculation to Simplify Multiprocessor Design - Sorin, al. (2004) (Correct)
No context found.
M. Thottethodi, A. R. Lebeck, and S. S. Mukherjee. SelfTuned Congestion Control for Multiprocessor Networks. In Proceedings of the Seventh IEEE Symposium on HighPerformance Computer Architecture, Jan. 2001.
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC