The State-of-the-art flow recognition techniques suffer from limitations of coarse-grained parallelims, large memory requirements, and limited throughput. The multi-core processor emerges as a promising approach for high-performance and scalalbe flow recognition. However, it imposes great challenges on parallelism and memory consumption. This project will study multi-core processor-based parallel flow recognition (pFlow), with particular focurses on parallel flow recognition architecture on the Multi-core platform, parallel signature matching algorithms and data structures, behavioral modeling and datamining, expeirmentation and evaluation. This project will improve the time/space efficiency of flow recogntion for network applications such as NIDS/NIPS, application-layer flow identifcation, and flow anomlay detection.

WP1-Flow Recognition Systems

  • Commodity hardware platform design and implementation.
  • Dedicate hardware platform design and implementation
  • High-level pipelined architecture for parallel flow recognition
  • Deployment of network virtualization over the hardware solutions
  • Implementation of openflow platform to export flow recognition results

WP2-Parallel Signature Matching Algorithms

  • Fine-grained parallel signature matching algorithms design
  • Time/space-efficient hash tables and Bloom filters for parallel signature matching

WP3-Behavioral Modelling and Data Mining

  • Sequence data mining
  • Automatic signature extraction
  • Efficient decision tree mining

WP4-Experimentation and Evaluation

  • Definition of evaluation scenarios and preparation of experiments
  • Validation and evaluation of high performance flow classification
  • Evaluation of network virtualization for high performance flow classification

WP5-Management and Dissemination