行為感知之TCP NAT穿透方法的選擇機制

Abstract

近年來，因為Network Address Translation (NAT) 的方便性與實用性，NAT被廣泛地應用在網路中。由於TCP在建立連線前會進行三向交握，此舉將增加NAT穿透的複雜度與難度。 因此本論文提出一套行為感知之TCP NAT穿透方法的選擇機制(Behavior Aware Selection Scheme, BASS) 用來穿透NAT以建立TCP連線。BASS的主要構想是欲連線雙方本機端(host) 的使用者代理人(user agent, UA) 先收集自己的網路拓樸環境和當地的NAT資訊，包含是否位於public domain下、NAT的mapping行為、filtering行為、是否支援hairpin與TCP狀態追蹤特性。接著BASS就可以利用此兩個UAs所收集到的資訊，選擇出一個最適合的NAT穿透方法，並通知這兩個UAs使用此方法進行連線測詴。因為BASS知道連線雙方TCP NAT traversal的支援能力，故可以避免使用不會成功的穿透技術來執行連線測詴，以減少連線測詴所耗費的時間與資源，更進一步判斷NAT是否具有hairpin特性和利用IP Spoofing方法來提升直連率。 我們已經完成BASS的實作，並且針對直接連線的比率、測詴時間與資源使用量等效能指標，進行兩類測詴。第一類是BASS與現有方法NBA的直連率比較，BASS的平均直連率為97.37%，明顯較NBA的平均直連率55.51%高。第二類是BASS與循序直連測詴(Sequential Connectivity Check, SCC) 以及平行直連測詴(Parallel Connectivity Check, PCC) 三種機制的效能比較。實驗結果顯示，在相同組合的NAT環境下，BASS和SCC與PCC的直連率相同，表示BASS選擇NAT穿透技術的方法非常準確，不會發生誤選的狀況。其次，當進行連線測詴的時候，BASS與SCC相較之下具有更短的測詴時間延遲，而且BASS比PCC使用更少的系統資源。故BASS的整體效能表現比NBA、SCC和PCC更為優異，可適用在現實NAT環境中。

In the recent years, due to the convenience and practicability of the Network Address Translation (NAT), NAT has been used extensively. Since TCP will use a three-way handshake before establishing the connection, the complexity of the TCP NAT traversal will be increased. In this thesis, we propose a Behavior Aware Selection Scheme (BASS) for TCP NAT Traversal Methods. With BASS, a user agent (UA) collects the network topology and the behaviors of the NAT device, such as public/private domain, mapping rules, filtering rules, hairpin support and TCP state tracking. When two UAs intend to establish a direct TCP connection session, the UAs report to a BASS server their respective NAT behavior information. The BASS server determines the most appropriate traversal method with the NAT information and then informs the UAs to check the connectivity through the selected method. Alternatively, the two UAs themselves can exchange NAT information with the help of a BASS server or a signaling server and determine the most appropriate traversal method. BASS can eliminate unnecessary checks, shorten the connectivity check delay, and reduce the number of message exchanges via utilization of the behaviors of the NAT devices. BASS considers the hairpin support behavior of NAT and uses IP Spoofing to increase DCR. We have implemented the BASS System and compared BASS with other schemes by evaluating three performance indexes: DCR, connectivity check time, and resource efficiency on real NAT devices. The result shows that BASS outperforms SCC in terms of latency and PCC in utilization of system resources when checking for connectivity. Furthermore, the average DCR of BASS is 97.37%, which is exceedingly higher than that of NBA’s, which is 55.51%.