重疊網路中多方串流服務之自治系統感知的中繼節點選擇機制

Abstract

本論文的目的是針對Overlay Network環境中的多人串流服務，提出一套以Autonomous System為基礎選擇中繼節點的機制，以利用網路中的節點來改變預設的網路繞送路徑，提升串流的傳遞品質。 本論文討論的Overlay Network一般稱為Peer-to-Peer (P2P)網路，由於P2P架構有許多優點，所以近年來有許多服務都採用P2P架構傳輸，例如：VoIP、視訊會議等。這一類的服務需要接近即時的傳輸，所以對串流傳遞的延遲時間有著非常嚴苛的要求，加上對畫面品質的要求，所以頻寬的要求也是考慮的因素之一。所以找出能夠符合服務需求的路徑，是一個很重要的議題。 目前所提出的解決方法是在現有的Overlay Network中，選擇一個中繼的節點幫忙繞送，亦即，來源端先送資料到中繼節點，再由中繼節點送到目的端。如此一來，便能夠改變繞送的路徑，進而提供較好的通訊品質。所以如何選擇適合的中繼節點，對傳輸的品質有很大的影響。目前有許多人提出選擇中繼節點的方法，但是目前的方法有許多的缺點，首先就是Scalability的問題，因為要監測所有的節點，所以無法在有太多節點的Overlay Network上運行。再者，這些方法選擇中繼節點的時間過長，會影響到服務的啟動時間。而且，目前的方法大都著重於兩人的通訊，並沒有針對多方通訊的特性來設計選擇中繼節點之機制。最後，目前的方法大多只有根據延遲時間或者是頻寬等的單一條件選擇中繼節點。 根據以上的觀察，本論文提出一套以Autonomous System為基礎的中繼節點選擇機制。這一套機制能夠解決先前所提到的問題。我們首先將加入Overlay Network的使用者，依照所在的Autonomous System來分群，利用相鄰節點狀態相近的概念，來減少對於中繼節點的Probe，接下來，在選擇中繼節點的時候，同時考慮頻寬以及延遲時間。同時，我們也利用同一通話使用者共用中繼節點的概念，來減少MCU的上傳頻寬。總結以上，本論文相較於其他方法的特色在於：(1)利用分群的概念，有效減少Probe的次數。(2)使用者共用中繼節點，減少MCU上傳之頻寬。(3)選擇中繼節點時，同時考慮頻寬和延遲時間的因素。 為了驗證我們所提出方法的可行性，我們使用了目前學術界中廣為人知的網路模擬器NS2來進行相關實驗和比較。由於目前並沒有針對多人服務所提出的中繼節點選擇機制，所以我們的方法會和沒有使用中繼節點的方法來進行比較，以確認使用中繼節點對網路傳輸的影響。實驗的結果顯示，我們所提出的方法能夠在不需花費太多時間在選擇中繼節點的情況下，有效的減少MCU的上傳頻寬，並且能夠支援更多的使用者。

In this thesis, we propose an effective Relay Node selection mechanism that utilizes the characteristics of Autonomous System (AS) for multiparty streaming service in Overlay Networks. With Relay nodes, we can change the default network routing path and get good quality of the streaming transmission. The Overlay Networks under discussion in this thesis are Peer-to-Peer (P2P) networks. P2P network architectures possess many salient features so that many real time multimedia services, such as VoIP and Video Conferencing, use P2P architectures as underlying transmission vehicles to convey streaming data. These services rely on the underlying network to transport real time data with strict delay requirements. Even more, Video Conferencing require the underlying network to provide more bandwidth to convey large volume of multimedia data. However, current P2P overlay can not fulfill the strict delay and high bandwidth requirement. Therefore, several researchers have proposed the use of relay nodes to resolve the above problems in Overlay networks. The use of relay may change the routing path and achieve better communication quality. Therefore, many researchers have proposed methods for relay selection. However, previously proposed methods have some drawbacks. Some of them have scalability problem, because nodes must monitor the statuses of many other nodes in an Overlay Network. Others may spend too much time on relay selection, resulting in long service start-up times. In addition, most of the previous methods focus on two-party communication, not on multiparty service, and consider either delay time or bandwidth only. In order to eliminate the above problems, we proposed a relay selection mechanism for multiparty services in Overlay networks. We first classify users by the AS a user belongs to. Because Internet adopts AS-based routing, that is, packets destined for a host in an AS are routed to the AS’s gateway first, a route to a host can represent the routes to other hosts of the same AS. As a consequence, we can reduce the probe messages because the results of probes to some hosts of an AS can be used as representative results of the probe to other hosts of the AS. We then propose a relay selection algorithm that considers both delay time and bandwidth. Furthermore, the relay selection algorithm also adopts the idea of in-use relay first to reduce the uplink traffic of MCU, and thus can support more users to join the session. In conclusion, our mechanism has following feature: (1) AS-based Relay node classification can reduce probe messages. (2) In-use Relay Selection reduces MCU uplink traffic. (3) Relay Selection Algorithm considers both delay time and bandwidth. We use NS2 network simulator to verify and evaluate our relay selection mechanism. We compare our mechanism with direct connection method because no any other Relay selection methods existing for multiparty service. Experimental results show that our proposed method spends little time on selecting relay nodes and can reduce the MCU upload traffic effectively to supports more users.