Randolph Y. Wang
ABSTRACT
In this paper, we study how to minimize the latency of a message through a network that consists of a number of store-and-forward stages. This research is especially relevant for today's low overhead communication systems that employ dedicated processing elements for protocol processing. We develop an abstract pipeline model that reveals a crucial performance tradeoff involving the effects of the overhead of the bottleneck stage and the bandwidth of the remaining stages. We exploit this tradeoff to develop a suite of fragmentation algorithms designed to minimize message latency. We also provide an experimental methodology that enables the construction of customized pipeline algorithms that can adapt to the specific system characteristics and application workloads. By applying this methodology to the Myrinet-GAM system, we have improved its latency by up to 51%. Our theoretical framework is also applicable to pipelined systems beyond the context of high speed networks.
- 1.Anderson, T., Culler, D., Patterson, D., and the NOW team. A Case for NOW (Networks of Workstations). IEEE M*cro (Feb. 1995), 54-64.]] Google Scholar
Digital Library
- 2.Anderson, T., Dahlin, M., Neefe, J., Patterson, D., Roselli, D., and Wang, R. Serverless Network File Systems. A CM Transactions on Computer Systems 14, 1 (Feb. 1996), 41-79.]] Google ScholarDigital Library
- 3.Boden, N., Cohen, D., Felderman, R., Kulawik, A., Seitz, C., Seizovic, J., and Su, W. Myrinet - A Gigabit-per-Second Local- Area Network. IEEE MICRO (Feb. 1995), 29.-36.]] Google ScholarDigital Library
- 4.Chun, B., Mainwaring, A., and Culler, D. Virtual Network Transport Protocols for Myrinet. in Proc. of 1997 Hot Interconnects V (August 1997).]]Google Scholar
- 5.Jacobson, V. pathchar - A Tool to Infer Characteristics of Internet Paths. http://www'msri'~rg/sched/empennage- /j acobson.html, 1997.]]Google Scholar
- 6.Jamrozik, H. A., Feeley, M. J., Voelker, G. M., II, J. E., Karlin, A. It., Levy, H. M., and Vernon, M. K. Reducing Network Latency Using Subpages in a Global Memory Environment. I# Proceedings of the Sixth International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS-VII) (Oct. 1996), pp. 258-267.]] Google ScholarDigital Library
- 7.Keleher, P., Cox, A. L., Dwarkadas, S., and Zwaenepoel, W. TreadMarks: Distributed Shared Memory on Standard Workstations and Operating Systems. In Proc. of the 1994 Winter" Usenix Conference (January 1994), pp. 115--132.]] Google ScholarDigital Library
- 8.Kent, C. A., and Mogul, J. C. Fragmentation considered tlarmful. In Proc. of Frontiers in Computer Communications Technology, A CM SIGCOMM (August 1987).]] Google ScholarDigital Library
- 9.Martin, R. P., Vahdat, A. M., Culler, D. E., and Anderson, T. E. Effects of Communication Latency, Overhead, and Bandwidth in a Cluster Architecture. In Proceedings of the Twenty-Fourth international Symposium on Computer Architecture (May 1997), pp. 85-97.]] Google ScholarDigital Library
- 10.Pakin, S., Lauria, M., and Chien, A. Itigh Performance Messaging on Workstations: lllinois Fast Messages (FM) for Myrinet. In Proc. of Supercomputing '95 (November 1995).]] Google ScholarDigital Library
- 11.Postel, .I. Internet protocol. Request for Comments 791, Information Sciences Institute, Sept. 1981.]]Google Scholar
- 12.Postet, J. Transmission control protocol. Request for Comments 793, Information Sciences Institute, Sept. 1981.]]Google Scholar
- 13.Prylli, L., and Tourancheau, B. New protocol design for high performance networking. Tech. Rep. 97-22, LIP-ENS Lyon, 69364 Lyon, France, 1997.]]Google Scholar
- 14.Thekkath, C. A., Mann, T., and Lee, E. K. Frangipani: A Scalable Distributed File System. In Proceedings of the A CM Sixteenth Symposium on Operating Systems Principles (Oct. 1997).]] Google ScholarDigital Library
- 15.von Eicken, T., Basu, A., Buch, V., and Vogels, W. U-Net: A User-Level Network Interface for Parallel and Distributed Computing. In {-'roc. of the 15th A CM Symposium on Operating Systems Principles (December 1995), pp. 40-53.]] Google ScholarDigital Library
- 16.yon Eicken, T., Culler, D., Goldstein, S., and Schauser, K. E. Active Messages: A Mechanism for Integrated Communication and Computatioll. In Proceedings of the Fifth International Conterence on Architectural Support for Programming Languages and Operating Systems (ASPLOS-V) (May 1992), pp. 256-266.]] Google Scholar
- 17.Yocum, K. G., Chase, J. S., Gallatin, A. J., and Lebeck, A. 1%. Cut-through delivery in trapeze: An exercise in low-latency messaging. In I'roc. of the Sixth IEEE International Symposium on High Performance Distributed Computing (August 1997).]] Google ScholarDigital Library
Index Terms
- Modeling communication pipeline latency
Recommendations
Modeling communication pipeline latency
In this paper, we study how to minimize the latency of a message through a network that consists of a number of store-and-forward stages. This research is especially relevant for today's low overhead communication systems that employ dedicated ...
Split-Path Enhanced Pipeline Scheduling
Software pipelining increases the loop execution throughput by overlapping the execution of successive iterations in a pipelined fashion. For loops with control flows, however, software pipelining is not straightforward because we need to consider the ...
Pipeline spectroscopy
ecs'07: Experimental computer science on Experimental computer sciencePipeline Spectroscopy is a new technique that allows us to measure the cost of each cache miss. The cost of a miss is displayed (graphed) as a histogram, which represents a precise readout showing a detailed visualization of the cost of each cache miss ...
Comments