第三層交換器訊務區分器之設計與實作

Abstract

訊務區分器在差異性服務中扮演很重要的角色。它不僅要根據IP目的地位址來決定下一個驛站，還需要有能力察看其他的欄位如來源位址、來源埠和目的埠來替供群址傳播，服務品質保證和其他加值性的服務如防火牆的功能。但是由於傳統的路由器是使用快取的架構來區分封包，這種架構沒辦法在任何時刻都有確定的性能，於是當網路流量很大時，就很容易形成頩頸。 在本論文中，我們強調最差情況時的狀況而不是採用平均的性能。我們提出一個綜合Crossproducting 和Lucent Bit Vector的方法。這個方法允許每個欄位以prefix, range或確定的值來設定。我們也把這個演算法實現在我們第三層交換器的雛形機中。它可以每秒處理一百萬個封包。

One of the most important elements to support differentiated services is the traffic classifier. It needs to be capable of parsing the destination address as well as other fields such as source address, TCP and UDP port numbers for supporting QoS, firewalls, and normal unicast and multicast forwarding. However, traditional routers are based on flow-cache architectures to classify packets. They can't provide predictable efficiency at all times and will easily become the bottleneck when the number of the active flows is extremely high. In this thesis, we emphasize worst-case performance over average case performance. A combination scheme that combines the advantage of Crossproducting and Lucent Bit Vector is proposed. It allows the classification rules to be based on multiple fields and each field to be able to specify ranges, prefixes, or exact values. This algorithm is also implemented into our layer 3 switching prototype, allowing the processing rate of 1 million packets per second.