利用內容定址記憶體的低耗能軟體定義網路封包分類機制

Abstract

在本篇論文當中，我們提出一個軟體定義網路交換機架構和封包分類方法以達到減少交換機運行耗能及硬體所占空間。單一三態內容定址型記憶體可以做平行搜尋，會比一般記憶體還要更快，因此被認為在接下來的高速網路時代極具潛力，但在運行時會耗去大量能量且比一般記憶體所占的空間還大。如果應用於軟體定義網路交換機，則又會需要儲存更大數量更大寬度的規則，則以上缺點又會更嚴重。因此，我們將單一記憶體分成兩塊較小型記憶體以分擔資料，並減少每次搜尋所開啟的規則數量，而我們也發現並不是一定需要所有欄位存在才能確保規則的獨立性，據此我們把一部分的資料存於固態隨機存取記憶體以減少內容定址型記憶體之負擔。此外，當控制台擁有一定的統計數據之後，可以根據封包的存取次數來決定其擺放的位置以得到比較高的存取優先權，如此可更進一步的改善每次搜尋所開啟的規則數量。我們有列出模擬結果以及計算結果也會和單一三態內容定址型記憶體做比較，其顯示我們提出的演算法的確能有效降低耗能及改善記憶體空間。

In this thesis, we proposed a SDN flow switch architecture and a packet classification method to reduce the searching power consumption and the occupation of board space. Ternary content addressable memories (TCAMs) can do parallel searching, they are faster than other memories to perform packet classification for high-speed network industry. But TCAMs have high power consumption and large board space, they also suffer from the well-known range expansion problem, if we use TCAMs to store the SDN flow tables which have more flow numbers and larger rule widths, the negative effects above become worse. To improve the shortages, we separate a single TCAM to one BCAM and one small TCAM to share the data loading, and reduce the number of entries charged per search. Besides, we found that it's un-necessary that a rule needs all fields to maintain its independent property in the table, so we put some of fields into SRAM words to reduce the loading of CAMs. When the SDN controllers get sufficient statistics, the classification method can decides rules' position of the CAMs based on the access rate of rules, these rules may get higher priority in the CAMs. Furthermore, we also show the simulation and numerical results, and compare with single TCAMs. The results show that the proposed method can efficiently improve power and space performance.