Our research interests include:
Our main interest in this field is to understand the statistical properties of the Web traffic in the Internet. To this end, we have analyzed several request traces at the server side and have shown that they exhibit self-similarity and long-range dependence properties. A report [1] (in French) is available. However, in contrast to the results reported in the literature, the Hurst parameter which measures the degree of correlation of the data, does not appear to be as large as for other data sets (including the Bellcore trace). This therefore suggests that Markovian traffic models (see [2] and [3]) should not be ruled out in the first place and may provide a good alternative to more sophisticated models featuring long-range dependence properties such as the model studied in [4].Another effort has been devoted to the statistical analysis of the frequency at which Web pages are modified. A prototype software tool has been developed to this end. This is still an ongoing work. Some preliminary results are reported in [5] (in French).
Caching at different levels is proposed and/or implemented in the Web: caching at the client side, in the backbone network and at the server side. As in traditional memory caching, a crucial question in Web caching is to decide which documents must be kept in the cache. In other words, when a new document must be brought into the cache and that this one is full (or heavily loaded), which document(s) should be removed from the cache? The strategy to make such decisions is usually referred to as caching policy. We have analyzed existing caching policies and have proposed new ones which turn out to be more efficient. Work have been done for both a proxy cache [6] and a server cache [7].
Our aim here is to develop both analytical and simulation tools for the performance evaluation of Web servers and Web applications. Currently we are investigating the performance models of the HTTP1.1 and the impact of different parameters used in the implementation of the protocol. We have been developing a software tool WAGON (Web trAffic GeneratOr and beNchmark) for generating Web traffic and for the performance evaluation of Web servers. Some results are available in [8] and [9].
We are maintaining a Web page on the literature related to the above mentioned fields. Any comments are welcome and can be mailed to Zhen Liu, Nicolas Niclausse and Cesar Jalpa.
Additional readings can be found in
- HTTP-NG Reading List
- PARC HTTP-NG: Web Characterization Reading List
- GVU's 7th WWW Survey Results
- Hints on Running a High-Performance Web Server
- Macintosh Webserver Performance
- LSAM Publications
- Cache Now
- Web caching architecture document
- Virginia Tech's Network Research Group
- WRL Research Technical Reports