A Scalable Carrier-Grade DPI System Architecture Using Synchronization of Flow Information

摘要

In this paper, the concept of deep packet inspection (DPI) is explained, and technical issues in deployment of DPI systems in a backbone network of Internet service provider (ISP) are mainly discussed. First, technologies of content identification using DPI are briefly explained, and types of clustering mechanisms that support a large bandwidth with a number of DPI systems are dealt with in detail. Second, a flow synchronization-based DPI architecture is introduced in order to achieve high scalability in system configuration and identification accuracy without rerouting of data flows. Based on the proposed architecture, flow information synchronization and adaptive traffic control using polling-based bandwidth arbitration and virtual queue synchronization are proposed, so that flows, which are randomly distributed over DPI modules by asymmetric routing, can be identified and controlled accurately. The proposed system architecture is practically manufactured as a 40-Gb/s system with four 10-Gb/s DPI modules, and performances of the proposed mechanisms are proved by practical tests that are executed using real traffic in an ISP backbone network. Finally, the scalability and the flexibility of the proposed mechanisms are reviewed from the viewpoint of enabling technologies of smart networks.