Traffic identification is an important issue in the Internet, especially where QoS and Service Level Agreements are enforced. Due to the enormous number of applications and protocols that run on non-standard port numbers, using metadata such as port numbers to identify Internet traffic is no longer practical. As an open-source effort, OpenDPI is a software package that identifies network flows based on Deep Packet Inspection technology by scanning packet payload for certain patterns. While OpenDPI achieves certain level of accuracy for the known protocols, its performance is highly limited by the time-consuming pattern matching processing. In this paper, we present our practice in migrating the software package to multi-core platforms which have been increasingly available to the computing world. The purpose of this effort is to improve the throughput of OpenDPI. Our experiments show that the performance gain is nearly linear to the number of cores on which OpenDPI runs, which demonstrates the scalability of our solution.
展开▼
