An analysis of decentralized peer-to-peer file sharing performance : An overview of how different parameters affect the average download time in a BitTorrent-like network

Sammanfattning: During the COVID-19 pandemic, online teaching has become more prevalent than ever before. As teaching resources such as video lectures are to be distributed from a single teacher to a large number of students, who may choose to download the files at any time of the day, questions arise as to how to do so efficiently. Traditionally, this may be done by uploading the file to a centralised server. This concerns not only economical aspects, such as renting the services of a centralised system, but also the consequences of depending on a third party providing such a service. A popular alternative is the decentralised peer-to-peer file sharing protocol known as BitTorrent. In this protocol, the peers download from each other as well as from the original downloader. In this protocol, the file is split into a number of pieces. Parameters such as the size of these pieces, the up- and download speeds of peers, as well as the file size, all affect the time it takes to download the files. To answer how these parameters affect the download time of torrents, a controlled environment was created with the capability to automatically run and control a large number of torrent clients. On this platform, 96 isolated practical experiments were run, each representing a unique combination of parameters. The result of these runs, as measured by average download time for the peers in each run, were compiled and presented as graphs. The results showed that the size of each piece in the file drastically affects the average download time, with larger pieces generally tending toward a significant increase in download times. Increasing the download speed resulted in a sublinear decline of download times, in large due to the upload speed being set to half the download speed, which turned out to be the actual limiting factor. Increasing the file size also did not linearly increase download times, and resulted in less drastic worsening of download times for larger piece sizes. The authors believe this to be in large due to larger file sizes resulting in the peers having more time to spread the pieces among each other, whereas smaller file sizes had to increasingly rely on the original downloader.