利用快取增加IPv6下的封包分類效能

Abstract

封包分類技術，是將封包依照其標頭的內容，將封包分類成不同的類別。封包分類目前應用在網路安全、網路服務分級、路由搜尋、流量分散等項目上。而目前的封包分類方式，可分為軟體及硬體實做兩種。而目前大部分的封包分類方式，在網路環境升級成IPv6後，由於標頭內容大幅增長，使得分類效能下降，整體網路效能大幅下滑。 在本篇論文中，我們提出使用快取方式，來增進封包分類在IPv6下的效率。我們對不同的快取記憶體容量、記憶體架構、以及快取內容的替換選擇方式等項目做了測試。在我們的實驗中，當快取記憶體的大小為1024格的情形下，快取的命中率可以達到9成以上，大幅增加封包分類的效率。

Packet classification is a technique that classifies flows into different class by the headers in packets. Packet classification is applied in network security, QoS, routing, load balancing, and etc. Currently, algorithms of packet classification are categorized into hardware or software solutions. Almost packet classification implementations can not be efficient enough when the network environment is upgraded to IPv6 because the much longer fields in the header. In this thesis, we proposed that use a cache memory to improve performance of packet classification in IPv6. We also simulated the performance under different cache sizes, architectures, replacement policies, and etc. According to the simulations, our schemes can achieve hit rates more than 90% when cache size is more than 1024 in 4-way associative cache memory architecture.