經由選擇性排程面對同儕網路系統中的連接限制

Abstract

在現今網際網路架構下，正面臨著網路位址不足的問題，此問題已由使用網路位址轉譯器暫時得到紓解。網路位址轉譯器在導引網路流量時，若是不知封包目的地，則會將連線阻擋起來，此現象在需要成員能夠接受外來連線要求的同儕系統中造成了系統效能的降低。此篇論文研究了使得同儕系統中能夠接受外來連線的成員先取得資料的方法，我們期望透過此方法，可以使得這些能夠接受外來連線的成員可以幫助同儕網路系統中資料的傳送。在檔案傳輸同儕系統中，我們的模擬呈現了百分之二十的效能增進。 此論文另一重點為考慮網際網路服務提供者(Internet service provider)的同儕系統傳輸設計。考慮到在同一網際網路服務提供者之下的傳輸效能較好，此篇論文的方法在資料傳輸排程時考慮資料傳送者的ISP，在模擬中，此方法於檔案傳輸以及多媒體串流的同儕系統均能大幅的減少跨網際網路服務者的網路流量，並且服務品質仍與原來未考慮ISP即進行排程時一致。

Recent study indicates that a large portion of peers in a P2P system do not contribute their out-going bandwidth due to the usage of network address translator (NAT) that blocks the incoming traffic. Recent study also shows that many internet service providers (ISPs) simply choose to block P2P connections because they create too much cross-ISP traffic. In this paper, we present a public-first approach to resolve the NAT-related connectivity constraint for P2P content delivery systems. Unlike STUN, which attempts to resolve the NAT traversal problem directly, our approach targets better utilization of available connectivity between peers and achieves this goal by selective content delivery scheduling. Experimental results show that it improves the P2P file transmission time by 20%. To reduce the cross-ISP traffic of P2P systems, we propose a light-weight, distributed method to identify peers of the same ISP. The proposed method reduces cross-ISP traffic without affecting the P2P system performance.