一個在同儕式網狀串流系統中之有效率的來源選擇方法

Abstract

目前大部分的網際網路服務皆採用傳統的用戶伺服器(client-server)模式，隨著網際網路的頻寬不斷擴展增加，需要較高網路頻寬的服務愈來愈被廣泛使用，例如，網路電話(Voice over Internet Protocol)、隨選視訊(Video on Demand)、網路電視(Internet Protocol Television)。然而，當使用者數目劇增時，用戶伺服器模式將遭遇到可容納度及高成本的問題。而在所有這些應用服務中，網路電視又是占用最多資源的。例如，頻寬、儲存空間，而且可能遭遇大量使用者同時使用所導致資源不足的問題。為了解決系統資源不足的問題，在近十年，發展出一種熱門的網路服務架構：同儕式網路(Peer-to-Peer Network)，簡稱P2P網路，越來越多的網路電視利用同儕式架構以提供服務。而因為P2P網路中，排程方法(Scheduling Method)影響到來源端到目的端所花的時間，每條連線的頻寬都不盡相同，如何有效的利用不同的頻寬，使得資料的下載時間可以縮短，讓使用者更快的取得所要的資料，以增進服務品質，是設計P2P排程方法的一大重點。 在本篇論文中，我們介紹一些現有的排程方法，並提出一個來源同儕(source peer)的選擇方法，利用一些探測封包，探測所有同伴(partner)的可用頻寬以及傳輸狀態，利用這些數值選擇出一個最佳的來源並從其取得資料，有效的利用狀態最佳的同儕，降低取得資料的時間。最後利用OMNet++模擬程式來驗證所提出的方法。

Many popular applications and services currently deployed on Internet are based on the traditional client-server model. As the internet steadily broadens the bandwidth scope, some application services such as VoIP (Voice over Internet Protocol), VoD (Video on Demand), IPTV (Internet Protocol Television) that required high bandwidth are deployed extensively. Although Client-Server model is intuitive and easy for implementation, however, the server will become a potential bottleneck when the number of users grows. It is always difficult for a server to provide the service to a large number of clients at the same time because of high cost and scalability problem. Among these services IPTV needs the largest amount of resources such as bandwidth and storage, and it may suffer shortage of resources when a certain number of users use the system at the same time. To accommodate system resource problem, a popular architecture - Peer-to-Peer (P2P) network for Internet service has been developed in recent decade, more and more IPTV services are deployed based on P2P architecture. In P2P network, data scheduling algorithm has a major effect on the data transfer time between the source and destination. Since every link has a different bandwidth, so how to efficiently use the different bandwidth to reduce the data downloading time so that users can get the data quickly becomes an important issue for designing the P2P scheduling algorithm. In this thesis, we proposed an efficient source peer selection algorithm which uses probing packets to detect both the available bandwidth and transmission status of all of its partners, and use these statistics to choose the current best source to retrieve data so that the download time can be minimized. Finally, OMNet++ simulation shows that our proposed approach can significantly reduce the downloading time with minimum overhead.