使用於多路徑傳輸控制協定之高效率按順序到達的調度演算法

Abstract

在現有網路環境中，多路徑的裝置無所不在，例如手機裝置中有 WiFi 與 3G/4G網路，而傳統 TCP 的設計只能建立一條路徑去傳送，而 MPTCP 則是改進傳統 TCP進而可以使用多條路徑去傳送資料。使用多條路徑傳送資料會因為不同路徑的延遲差異造成封包得隊頭阻塞(Head-Of-Line Blocking)，在接受視窗小的時候，隊頭阻塞的問題會影響整體的吞吐量。為了避免此問題，我們嘗試讓封包按照順序地抵達接收端，好讓隊頭阻塞的問題大大減少。在 Tuan-Anh Le 與 Loc X. Bui 的研究裡，讓 MPTCP排程器選擇不連續的序號發送使接收端收到連續的封包，在此論文中，我們發現此演算法在兩條路徑以上是不良善的，需要修正論文中的序號算法。此外，為了整體吞吐量，我們加入了接受視窗的考量，在犧牲一點亂序的封包情形下，以追求最大吞吐量，從模擬的結果我們改善此篇論文的缺失，且比此篇論文用更少的接收視窗就可達到最大的吞吐量。

In the modern network, multihomed devices are omnipresent. For example, a mobile phone has Wi-Fi network and 3G or 4G network. Nevertheless, the original TCP transport the data on just only one path. MPTCP is the protocol that evolves the original TCP in order to use multiple paths for transportation. Using MPTCP may cause Head-Of-Line Blocking problem due to the differential delay of paths. With a smaller receive window, the Head-Of-Line Blocking problem may affect overall throughput. To reduce this problem, we attempt to let packets arrival in-order at receiver so that the Head-Of-Line Blocking problem greatly diminished. In the research of Tuan-Anh Le and Loc X. Bui, the scheduler sends non-in-order packets so that the packets arrive in-order at receiver. In their paper, we find that the algorithm is incomplete for more than two paths and needs to be modified in index formula. Additionally, we consider the receive window for the purpose of overall throughput. With sacrificing the in-order rate, we pursue the maximum of the throughput. From our simulation, our method improves the reference paper's flaw and use less buffer to get maximum throughput than it uses.