Self-similarity in World Wide Web traffic: evidence and possible causes

Abstract

Recently the notion of self-similarity has been shown to apply to wide-area and local-area network tra c. In this paper we examine the mechanisms that give rise to the self-similarity of network tra c. We present a h ypothesized explanation for the possible self-similarity of tra c by using a particular subset of wide area tra c: tra c due to the World Wide Web (WWW). Using an extensive set of traces of actual user executions of NCSA Mosaic, re ecting over half a million requests for WWW documents, we examine the dependence structure of WWW tra c. While our measurements are not conclusive, we show evidence that WWW tra c exhibits behavior that is consistent with self-similar tra c models. Then we show that the self-similarity i n s u c h tra c can be explained based on the underlying distributions of WWW document sizes, the e ects of caching and user preference in le transfer, the e ect of user \think time", and the superimposition of many such transfers in a local area network. To do this we rely on empirically measured distributions both from our traces and from data independently collected at over thirty WWW sites.

A Dynamic Timeout Control Algorithm in Software Defined Networks

“…2 (a). Using the self-similarity properties [8], [12], [13], [14], we can further notice that the inter-arrival times of packets that belong to the same flow also follow the same distribution as shown in Fig. 2 (b).…”

Benchmarking Models and Tools for Distributed Web-Server Systems

“…The generation of synthetic Web traffic is not a trivial task because it aims at reproducing as accurately as possible the characteristics of real traffic patterns, which exhibit some unusual features such as burstiness and self-similarity [4,12]. On the other hand, real world workloads are inherently irreproducible, since it is impossible to replicate the overall conditions under which the performance testing was originally performed.…”

“…Time characteristics are typically not taken into account, hence the stream of requests depends only on the filelist while the inter-arrival request time is set. The filelist approach lacks of flexibility with respect to the workload specification, and also ignores the concept of user sessions As discussed in [4,3,8,12], Web traffic is bursty, sessionoriented, and characterized by heavy-tailed distributions, which have high or even infinite variance and therefore show extreme variability on all time scales. To emulate these workload characteristics, it is not sufficient to mimic the user activity by requesting a set of files as quickly as possible; it is necessary to provide some support for modeling the session-oriented nature of Web traffic.…”

“…Therefore, the user behavior is modeled as a bursty two-state ON/OFF process, where ON periods correspond to the transfer of Web objects, and OFF periods correspond to the silent intervals after that all objects in a Web page have been retrieved. It has been demonstrated that the superposition of a large number of ON/OFF sources results in self-similar traffic, if the durations of ON and OFF phases are described by heavy-tailed distributions [12,43]. The characteristics of the request stream are specified through heavy-tailed distributions as regarding file size, request size, file popularity, embedded object references, temporal locality, and OFF times.…”

The use of a non-integer order PI controller with an active queue management mechanism

“…Therefore, the evolution of the window for these streams is identical. (Crovella and Bestavros, 1997;Karagiannis et al, 2004;2014b;. These features have great impact on network performance (Domańska and Domański, 2005;Domański et al, 2008;Domańska et al, 2012).…”