BitTorrent在多檔案下載環境中改進方法之研究

Abstract

BitTorrent (BT)是一個廣泛被使用的點對點(peer-to-peer)檔案傳輸協定。近年來，BT的使用者通常會同時下載多個檔案，而不是一次只下載單一檔案；我們稱之為：多檔案下載。在多檔案下載環境中，BT的使用者可以使用不同的下載方式，來安排這些檔案的下載；例如使用多個session來進行多個檔案下載(multi-session下載方式)，或是只使用一個session來下載所有檔案(single-session下載方式)。此外，目前BT採取了平均分配的策略，將使用者的上傳頻寬平均分配給它的上傳對象；但是我們認為，一個好的上傳頻寬分配策略，應該要隨著上傳對象的情況來動態調整使用者的上傳頻寬。我們的研究目的是找出在多檔案下載環境中最適合的下載方式；並且設計不同於以往採取平均分配的上傳頻寬分配策略，來改進使用者與整個系統的效能。 我們的研究發現，BT使用者應該在多檔案下載環境中應該採取一種single-session (expanded)的下載方式，以達到較好的下載效能。此外，我們也提出了三個上傳頻寬分配策略：reciprocal、utilization以及fairness策略。在我們的實驗中發現，reciprocal和utilization策略可以改進使用者與系統的下載效能，而fairness策略則可以增進系統的公平性。

BitTorrent (BT) is a widely used peer-to-peer (P2P) file sharing protocol. Recently, BT peers usually download multiple files simultaneously, rather than download a single file. We called this as the 'multi-torrent downloading'. In multi-torrent downloading, BT peers can use different ways to arrange the downloading of their files. For example, a peer can use multiple sessions to download multiple files (multi-session downloading); or it can put all the files together, and uses a single session to download all these files (single-session downloading). Moreover, present BT peers apply the equal-allocation strategy for allocating their upload bandwidth. That is, a peer will equally allocate its upload bandwidth to the uploaded targets. However, we think that a good upload bandwidth allocation strategy should dynamically adjust the upload bandwidth allocation depending on the situation of the uploaded targets. Our goal is to find a good downloading way in multi-torrent downloading environment; and we are going to find new upload bandwidth allocation strategies to improve the performance of peers and the system. Our research found that peers should use the single-session downloading (expanded) in multi-torrent downloading to achieve better downloading performances. We also proposed three upload bandwidth allocation strategies (reciprocal, utilization and fairness strategy). The reciprocal and utilization strategy improve the downloading performance of peers and the system; the fairness strategy improves the system fairness.