The emerging of content providers such as YouTube induces a rapidly increasing demand for multimedia streaming, which augments the network resource consumption. Current content distribution networks however are not suited for the high quality video streaming we can expect in the future. Nevertheless, Peer-to-Peer (P2P) networking ensures a high degree of scalability and is a possible solution.On the other hand, P2P also imposes higher resource requirements on the end users. The end users have to use disk access time and CPU resources in order to serve requests. This resource consumption can reduce the playback quality, and it is therefore desirable to reduce the resource cost of P2P networking. One method is to employ caching.
While P2P networking seems promising with respect to scalability issues, it also creates traffic patterns that make current caching strategies insufficient. This thesis examines different caching techniques and their performance with P2P traffic patterns. These differ from regular patterns since clients request individual blocks of a file from multiple providers, instead of downloading the file as a whole from one provider alone.
In this thesis we show that existing caching algorithms are efficient in combination with P2P multimedia streaming. Multiple difficulties associated with P2P traffic patterns have been detected. To solve these problems, we propose a new and improved caching technique called Relevance Based Caching (RBC). RBC uses prefetching, in which the most relevant blocks are cached. The caching algorithm identifies the P2P access pattern, and together with the popularity of the individual blocks and the files as a whole, it calculates relevance values for each block. We show that by using this algorithm, we obtain a good performance, without exerting too high resource demands on the end users.