Behavior and Performance of the Packet Count Flow Classifier

摘要

Flow classification is used in the Internet to find and detect flows of packets that can be handled separately from the rest of the traffic. The packet flows may receive different service levels or they may be treated differently in the network. This work studies a flow classification method where the flows are detected by counting packets on flow candidates. This counting is continued until the packet count threshold is reached and an active flow is established. This work studies the packet classifier using a set of traffic measurements made in different networks. Results show that the number of packets outside of any active flows is relatively small thus reducing the workload of conventional packet forwarding. However, the classified flows are from a large number of different applications and it is doubtful that packet count flow classification may be used to support the introduction of quality of service in the Internet. Furthermore, the candidate table, an essential part of the packet count classifier implementation, forms a significant performance bottleneck because of the size and frequent changes in it.