Research And Implementation Of Heterogeneous Parallel Flow Classification Mechanism For EMD

The efficient flow classification strategy is a prerequisite and guarantee for flow control and management. With the rapid rate of network data upgrade and network operations are becoming increasingly diverse. Flow classification has become one of the bottlenecks to improve router performance. Based on the national "863" plan's major special topic "Research on system performance and key technologies of large-scale Access Convergence Router", Through the in-depth study of the flow classification strategy for complex network applications environment and user environment, design and implement a new efficient heterogeneous parallel flow classification mechanism, which is effective to solve the high-speed flow's multi-dimensional classification problems. Just as the test results show: this flow classification mechanism performe well in EMD (Ethernet Multiplexer-Demultiplexer) system. This flow classification mechanism can fully meet the functional requirements and performance requirements in EMD system.Concretely, this paper includes the following aspects :1) Based on the study of flow classification, we give a formal mathematical description of the flow classification problems. By comparing and analyzing the existing classification algorithms systematically, we point out the limitations of existing algorithms.2) In EMD system, the flow classification technology must support multi-dimensional and high-speed flow classification, aim at this, we present a heterogeneous parallel flow classification mechanism. Through parallel processing, solve the bottleneck of wire-speed processing of high-speed data flow. By designing each sub-layer's classifier, system achieves the rationalization of the application of the hardware resources and performance optimization of different rang match, the flow classification mechanism not only fully meet the EMD system's function and performance requirements, but also has good scalability. It can be used to guide high-speed, multi-service network's flow classification design3) Analyse the network authentication flow characteristics, In EMD system, the user authentication flow's filter set is smaller, and such IP address can be accessed without authentication. Aim at such characteristics. We present a flow classification strategy for user authentication flow. In this strategy, the TCAM hardware resources will not be occupied, just only use little FPGA resources, it can meet the 10Gbps wire-speed process demand of user authentication flow, it is also easily to be implemented. The simulation shows that: this strategy can fully meet the system performance requirements.4) In EMD system, some flow classification action is high speed need and its table items structure is complex, aim at such demands, we present a TCAM-based pipeline flow classification, the strategy not only gets higher throughput than ordinary TCAM flow classification strategy, but also gets better balance between TCAM bandwidth utilization and FPGA resources. Analysis shows that TCAM-based pipeline flow classification strategy effectively meet the performance requirements of complex flow classification operation.5) Based on Xilinx Virtex4 LX160 (FPGA). IDT71T75602 (SRAM), IDT75K72100 (TCAM) devices. Take full advantage of hardware high search speed, simplicity features. use FPGA, TCAM and SRAM to implement heterogeneous parallel flow classification engine. Test results show that the heterogeneous parallel flow classification mechanism which is presented in this paper is able to satisfy the classification demands for EMD.