Traffic locality characteristics in a parallel forwarding system

Due to the widening gap between the performance of microprocessors and that of memory, using caches in a system to take advantage of locality in its workload has become a standard approach to improve overall system performance. At the same time, many performance problems finally reduce to cache performance issues. Locality in system workload is the fact that makes caching possible. In this paper, we first use the reuse distance model to characterize temporal locality in Internet traffic. We develop a model that closely matches the empirical data. We then extend the work to investigate temporal locality in the workload of multi-processor forwarding systems by comparing locality under different packet scheduling schemes. Our simulations show that for systems with hash-based schedulers, caching can be an effective way to improve forwarding performance. Based on flow-level traffic characteristics, we further discuss the relationship between load-balancing and hash-scheduling, which yields insights into system design. Copyright © 2003 John Wiley & Sons, Ltd.

[1]  Virgílio A. F. Almeida,et al.  Characterizing reference locality in the WWW , 1996, Fourth International Conference on Parallel and Distributed Information Systems.

[2]  J. G. Fletcher,et al.  An Arithmetic Checksum for Serial Transmissions , 1982, IEEE Trans. Commun..

[3]  Richard G. Baraniuk,et al.  Connection-level analysis and modeling of network traffic , 2001, IMW '01.

[4]  Anja Feldmann,et al.  Fitting mixtures of exponentials to long-tail distributions to analyze network performance models , 1997, Proceedings of INFOCOM '97.

[5]  Gero Dittmann,et al.  Network Processor Load Balancing for High-Speed Links , 2000 .

[6]  Jeffrey R. Spirn,et al.  Program Behavior: Models and Measurements , 1977 .

[7]  D. C. Feldmeier,et al.  Improving gateway performance with a routing-table cache , 1988, IEEE INFOCOM '88,Seventh Annual Joint Conference of the IEEE Computer and Communcations Societies. Networks: Evolution or Revolution?.

[8]  Joel L. Wolf,et al.  Synthetic Traces for Trace-Driven Simulation of Cache Memories , 1992, IEEE Trans. Computers.

[9]  Raj Jain,et al.  Packet Trains-Measurements and a New Model for Computer Network Traffic , 1986, IEEE J. Sel. Areas Commun..

[10]  Allen B. Downey,et al.  Evidence for long-tailed distributions in the internet , 2001, IMW '01.

[11]  David Thaler,et al.  Using name-based mappings to increase hit rates , 1998, TNET.

[12]  Sally Floyd,et al.  Wide area traffic: the failure of Poisson modeling , 1995, TNET.

[13]  Raj Jain,et al.  A comparison of hashing schemes for address lookup in computer networks , 1992, IEEE Trans. Commun..

[14]  Paul Barford,et al.  Generating representative Web workloads for network and server performance evaluation , 1998, SIGMETRICS '98/PERFORMANCE '98.

[15]  Jean Calvignac,et al.  IBM—PowerNP Network Processor , 2003 .

[16]  Craig Partridge,et al.  A Fifty Gigabit Per Second IP Router , 2001 .

[17]  Yitzchak M. Gottlieb,et al.  Building a robust software-based router using network processors , 2001, SOSP.

[18]  Ellen W. Zegura,et al.  Performance of hashing-based schemes for Internet load balancing , 2000, Proceedings IEEE INFOCOM 2000. Conference on Computer Communications. Nineteenth Annual Joint Conference of the IEEE Computer and Communications Societies (Cat. No.00CH37064).

[19]  David Waitzman,et al.  A 50-Gb/s IP router , 1998, TNET.

[20]  J. Le Boudec,et al.  Adaptive load sharing for network processors , 2002, Proceedings.Twenty-First Annual Joint Conference of the IEEE Computer and Communications Societies.