IEEE 802.11 無線網路和SIP-based VoIP交遞之實驗與分析

Abstract

本論文針對現有802.11無線網路及部署環境，利用SIP-based的VoIP軟體，分析現有無線區域網卡和微軟Windows系統的交遞 (handover)行為以及交遞過程中跨階層的交互影響，並逐層探討與歸納這些行為所造成的各項延遲。在無線網路的環境裡，當行動節動在漫遊時，偵測到訊號品質降低而影響通訊穩定度時，為了能保持在通訊狀態，行動節點會重新找尋目標基地台(Access Point, AP)，並與選定的目標基地台重新建立連線，這種交換基地台的過程稱為鏈路層 (link layer) 交遞。交遞的行為包含目標基地台探索(probe)、認證(authentication)與連結 (association)等三個步驟。 Mishra等人[3] 在無線區域網路鏈路層交遞的實證研究中，指出基地台的探索是無線區域網路交遞程序中最耗費時間的步驟；但卻未探討無線區域網卡的實作與網路部署環境對鏈路層交遞所造成之延遲，也未分析鏈路層交遞對上層通訊協定行為的影響。各廠牌的無線區域網卡因設計思維所著重的層面不同，因此所產生的交遞延遲也有差異；在更壞的情況下，若行動節點所連線的目標基地台與原服務基地台分屬不同的網域，則鏈路層交遞會進一步觸發網路層的交遞，導致整個交遞時間過長。以網路語音(Voice over Internet Protocol, VoIP)即時應用軟體為例，過長的交遞時間可能會干擾甚至中斷現有的語音通訊連線，影響通話品質。而且，若應用軟體本身對網路連線中斷是敏感的，則應用軟體需在網路連線恢復後與通訊對方重建應用層的通訊連線，這也意謂著延遲時間將拉的更長。因此，在本篇論文裡，我們透過有系統的方式，逐層分析及探討不同廠牌的無線區域網卡、不同的網路環境組態以及應用軟體實作上的差異的交遞行為，以及在交遞過程中跨階層的交互影響。 在這些實驗中，我們發現許多有趣現象。(一)鏈路層的交遞偵測(handover detection)對交遞延遲是有顯著影響的，且各廠牌無線區域網路卡設計的偵測方法也不相同。(二)無線區域網卡於交遞前後所連結的基地台屬於不同之ESS，稱之為inter-ESS；各廠牌無線區域網卡在基地台探索行為雖不相同，但在交遞的過程裡，大都傾向尋找原基地台所屬的的延伸式服務組合 (Extended Service Set, ESS) 的基地台與之連線，因此當行動節點進行跨ESS漫遊時，這種保持同一ESS的特性，會導致較長的基地台探索延遲。(三)不同無線區域網路組態管理工具(例如Windows Zero Configuration及Intel PROSet)，對基地台的探索行為也會產生影響。(四)當無線區域網路卡對所要連結的基地台下達連結請求時，普遍的實作方法仍會對所要連結的基地台送出探索訊息後，才會進行連結，這樣的作法比較沒有效率，尢其是當無線區域網卡不支援單點探索(unicast probe)時，基地台的探索會花費更長的時間。(五)鏈路層事件觸發(event triggers) 若沒有伴隨充足的資訊給上層通訊協定參考，鏈路層事件觸發會有正反兩面的效果。(六) 相對節點上的應用程式若有能力收送行動節點新IP位址的資料，也可縮短整體的交遞延遲。

In this thesis, we present the experimental results and analysis of the handover behaviors of the IEEE 802.11 wireless network (WLAN) and the cross-layer interaction between WLAN adapters and Microsoft Windows operating systems during handoffs using SIP-based VoIP applications. In wireless networks when a mobile node (MN) detects poor link performance (e.g., low received signal strength or signal-to-noise ratio, or high frame error rate), the MN may have to change its point of attachment to the Internet from one access point (AP) to another in order to retain its connection. The link-switch process is called a link layer handover, and involves AP probe, authentication, and association phases in 802.11 WLANs. Mishra et al. have conducted an empirical study on link layer handover and shown that probe delay constitutes most of the link layer handover latency. However, they did not explore the handover delays caused by different WLAN adapters implementations and network deployments, and the effects of the link layer handovers on the behaviors of the upper layer protocols.. The link layer handover behaviors may vary with different WLAN adapters, resulting in different handover delays. Even worse, a link layer handoff may further cause a layer-3 network layer handover if the new AP is situated in another subnet. Such handover delays may disturb or even break an ongoing session, especially for real-time services, such as Voice over Internet Protocol (VoIP) applications, Furthermore, the implementation of the application software itself also affects the handover delays. Therefore, in this thesis, we conduct a layer-by-layer analysis of handovers and investigate, systematically, the cross-layer effects of WLAN adapters, network configurations and application software on handover delays. Our experimental study discovers several interesting results. First, link layer handover detection has a significant effect on handover delays and different WLAN adapters adopt different heuristic approaches to link layer handover detection. Second, different WLAN adapters employs different probe behaviors, and most WLAN adapters first look for the APs in the same extended service set (ESS) as the AP the adapter currently associates with during handovers. The adherence to the same ESS may contribute a significant amount of probe delay when a WLAN adapter makes an inter-ESS movement, that is, from the coverage area of an AP to the one of another AP in a different ESS. Third, different configuration tools, such as Windows Zero-configuration or Intel PROSet, of WLAN adapters have different probe behaviors. Fourth, the probe-before-associate of WLAN adapters makes direct connection inefficient, especially when WLAN adapters do not support unicast probes. Fifth, the most interesting observation, link layer event triggers have both their pro and con if not accompanied by sufficient information. Finally, the capability for applications to receive/send data from/to the new IP address of a corresponding node will shorten the overall handover latency.