Special Issue on Network Traffic Measurements and Workload Characterization Carey Williamson Department of Computer Science University of Saskatchewan carey@cs.usask.ca This special issue of Performance Evaluation Review (PER) is devoted to Network Traffic Measurements and Workload Characterization. The topic is a timely one, as Internet traffic volume continues to grow, roughly doubling every 6 to 12 months. This growth is due in large part to the immense popularity and prevalence of the Web. Understanding the nature of Internet/Web traffic flows is important to researchers, protocol designers, network engineers, performance modelers, service providers, and capacity planners. This issue contains four interesting papers, each featuring the presentation and analysis of empirical measurements from operational networks. The measurements focus primarily on Web, IP, and Internet traffic workloads in commercial, academic, and wide-area network environments. While each paper differs in terms of the measurement tools and environment studied, all four papers offer insights into the present (and future) demands of Internet Web traffic. The first paper is by Judith Jerkins and Jonathan Wang from Telcordia Technologies (formerly Bellcore) and John Monroe from Southwestern Bell. This paper presents measurements of Internet traffic being carried over a telecommunications network using Frame Relay technology. Results are presented at both the framelevel and the call-level. Frame-level analyses highlight the asymmetry of many Internet traffic flows, the long-range dependence (LRD) in the frame-count and byte-count processes, and the impacts of coarse-grain link-level load-balancing on overall traffic structure. Call-level analyses show that session durations are heavy-tailed, and that the session arrival process is Poisson, with time-varying rate. The second paper, by Raja Epsilon, Jun Ke, and Carey Williamson from the University of Saskatchewan, presents IP/ATM traffic measurements from a commercial Internet Service Provider (ISP) environment. Traffic analyses at the cell-level, frame-level, and TCP connection-level highlight the bursty and asymmetric nature of Web traffic flows, and a trimodal distribution of frame (packet) sizes. Results are also presented for the distributions of bytes, packets, and time duration for Web TCP connections. The third paper is by Martin Arlitt, Rich Friedrich, and Tai Jin from HP Research Laboratories in Palo Alto. This paper uses five months of access log data from a Web proxy server to study Web workloads in an environment where clients have high-speed modem access. Results indicate larger mean, median, and maximum transfer sizes than reported in earlier Web workload studies, as well as steady monthly growth both in traffic volumes and in number of users. The final paper, by Paul Barford and Mark Crovella of Boston University, describes the design of a wide-area network infrastructure for Internet/Web performance measurements. The advantage of this new infrastructure is the simultaneous collection of client, server, and network-level performance data, using a combination of active and passive measurement techniques. The paper presents preliminary results from this measurement infrastructure, highlighting some surprising observations regarding the relationship between server and network performance. The authors of this paper are currently seeking other host sites for their measurement infrastructure. Please contact them if you are interested. I hope that you enjoy this special PER issue on network measurements and workload characterization. I certainly enjoyed putting it together. I sincerely thank all the authors for their diligent work in preparing papers for this issue, and for their timely responses to my editorial suggestions. I would also like to thank Scott Leutenegger for suggesting this topic, and for his patience and assistance with production of this issue. Happy reading!

/lp/association-for-computing-machinery/special-issue-on-network-traffic-measurements-and-workload-umBYbK8IiA