Modeling the data-dependent performance of pattern-matching architectures

  • Authors:
  • Christopher R. Clark;David E. Schimmel

  • Affiliations:
  • Georgia Institute of Technology, Atlanta, GA;Georgia Institute of Technology, Atlanta, GA

  • Venue:
  • Proceedings of the 2006 ACM/SIGDA 14th international symposium on Field programmable gate arrays
  • Year:
  • 2006

Quantified Score

Hi-index 0.00

Visualization

Abstract

There has been a significant volume of recent work on reconfigurable designs for pattern matching at high data rates with large pattern sets. Although various pattern-matching architectures and implementations have been presented, attempts to compare different designs have been inconclusive, or even misleading, due to variations in testing procedures. There has been no general methodology for analyzing and comparing pattern-matching system design approaches. In this paper, we present a software toolset and an associated methodology that is then used to evaluate pattern-matching circuit performance. This evaluation framework relies on an analytical model of FPGA pattern-matching architectures that quantitatively expresses the relationships between pattern properties, circuit area, and circuit delay. Using our model and toolset, we show how the efficiency and performance of each architecture is dependent on certain properties of the pattern set. A number of experiments are performed to demonstrate that the model does indeed accurately represent the area and delay in FPGA implementations of the given architectures. For several architectures and multiple pattern sets, circuit netlists are generated and then compiled for the Xilinx Virtex II Pro platform. Our results show that pattern set properties such as pattern length and alphabet size impact the circuits of each pattern-matching architecture in markedly different ways. This indicates that useful insights are gained by using this model-based analysis methodology.