Abstract
The main problem we explore in this paper involves predicting the performance of Web-resource downloads from unknown Web servers, based on knowledge about client-to-unknown-server network paths and performance measurements carried out on the set of known Web servers. We propose unknown-server-to-known-server topology-aware distance metrics based on the knowledge of network paths to both unknown and known servers at the autonomous systems level of Internet organization. The throughput value we want to predict for an unknown-server is approximated by the value achievable for the known-server—called the best one—with the least value of unknown-server-to-known-server distance metrics. The best server is selected using the nearest neighbor algorithm. The usefulness of this method for Web-performance prediction has been confirmed in real-life experiments. The results of the work allowed us to formulate positive recommendations for applying this approach to efficient gaining of Web resources in replicated content systems, file mirrors, content delivery networks, and digital libraries.
This work was supported by the Polish Ministry of Science and Higher Education under Grant No. N516 032 31/3359 (2006–2009).
The subject matter of this paper is covered by pending patent applications at the Patent Office of the Republic of Poland.