用於網路封包分類器的特殊用途內容定址記憶體

網際網路的形成是由網狀的連接點相互連接而成，而這些網路點則是由路由器(router)構成。由於網際網路資料量的成長迅速，造成現有的頻寬已不敷使用，因此衍生出差異性服務的概念。差異化的服務則是需要由每個網路傳輸點上的每個路由器查看封包上網路協定的標頭，並且給予所需要的動作，資料才能夠依照管理者所定義的目標順利的傳送至目的地。資料要進行分類時，需要去查看標頭上的資料與先前定義的規則是否相符合，一但規則數目增加，則查詢時所耗費的時間也會隨之增加。當內容定址記憶體的提出後，由於特殊的硬體架構設計，可以在固定和快速的時間內完成查詢的動作，這樣的結果使得內容定址記憶體非常適合用於高速的封包分類器的查詢動作。內容定址記憶體需要較多成本和功率上的消耗，這樣會使得使用上時受到限制。資料搜尋的方式上是使用以位元為比對的方式，這樣比對方式在最差情況之下會造成(2W-2)的倍乘因子(multiplicative factor)，一條規則可能會需要數個內容定址記憶體的內容儲存，使得原本成本就很高的內容定址記憶體的使用更是沒有效能。因此在本論文中提出了一個特殊用途內容定址記憶體，由於特殊的設計使得不會產生倍乘因子，進而一個規則只使用到一個特殊用途內容定址記憶體的內容儲存，不會使得成本再提高。對於未來的網路協定IPv6的到來，則可以使用以特殊用途內容定址記憶體為基礎的組合架構完成封包分類所需的資料搜尋。最後，為了實現特殊用途內容定址記憶體，使用了TSMC 0.35μm的製成以Fully-Custom的方式實現，並且使用電晶體層次模擬的軟體TimeMill進行功能和時序上的驗證。

關鍵字

內容定址記憶體；封包分類器

並列摘要

The internet is formed by reticulated connections that are mutually connected and these network points are composed by a router. In virtue of the Internet information data is increasing rapidly makes the existent bandwidth not useful enough. Therefore brings the concept of different service. Different service needs that routers in each network point look up the internet protocol in the packet header and give a suitable action at the right time. Then the data can be sent to the destination without any problem according the administrator. When the data is being classified needs to look up the information in the header and find out if they can fit in with the rules defined previously. Once the quantity of rules increases, the look up time increases proportionally. The special hardware architecture design of a Content addressable memory can finish the look up in a specific time which makes it suitable to look up in a high speed packet classification. But the use of it is limited because the Content addressable memory needs more cost and power consumption. The data search is compared by bit, therefore could produce a (2W-2) expand factor as the worst situation. One rule probably needs a number of Content addressable memory entries which makes it no efficient. This paper proposed a special purpose content addressable memory to avoid the multiplicative factor where one rule only needs an entry. The coming of IPV6 can use the special purpose content addressable memory to compose the frame that the packet classification needs for data search. Finally we use the TSMC 0.35um and the fully-custom methodology to implement the special purpose content addressable memory. Then we use transistor level simulation tool called “TimeMill” to verificate the functions and the timing.

並列關鍵字

Content Addressable Memory ； Packet Classifier

參考文獻

[3]P. Gupta and N. McKeown , “Algorithm for Packet Classification”, IEEE Network , Volume: 15 Issue: 2 , pp. 24 –32, 2001.

[4]A. Feldman and S. Muthukrishnan, “Tradeoffs for Packet Classification,”Proc. INFOCOM, vol. 3, Mar. 2000, pp. 1193–1202.

[5]V. Srinivasan et al., “Fast and Scalable Layer four Switching,” Proc. ACM Sigcomm, Sept. 1998, pp. 203–14.

[8]V. Srinivasan, S. Suri, and G. Varghese, “Packet Classification using Tuple Space Search,” Proc. ACM Sigcomm, Sept. 1999, pp. 135–46.

[9]T.V. Lakshman and D. Stiliadis. “High-speed policy-based packet forwardingusing efficient multi-dimensional range matching”, Proceedings of ACM Sigcomm, pages 191-202, September 1998.

國際替代計量

用於網路封包分類器的特殊用途內容定址記憶體

未授權

主題瀏覽