Generalized Pattern Matching Micro-Engine

Yuanwei Fang; Raihan Rasool; Dilip Vasudevan; Andrew A. Chien. 25 March, 2014.
Communicated by Andrew Chien.


Important applications including dictionary-based decoding, deep packet inspection, Bioinformatics (DNA Alignment), and JSON/XML processing depend on fast pattern matching. However, such applications are hard to accelerate. We explore a novel heterogeneous architecture to accelerate such FSM-based applications, balancing programmability and performance. The Generalized Pattern Matching micro-engine (GenPM) includes a novel micro-architecture, and software interface. We implement and evaluate GenPM in a 32nm TSMC process using a Snort network monitoring workload. Results show 8-wide16-step GenPM achieves 200x reduction instruction count and >200x performance increase. And more aggressive designs with greater width can deliver as much as 1700x performance improvements. Energy efficiency benefits range from 13x to 516x. Comparison show that GenPM improves performance and energy efficiency dramatically, approaching ASIC efficiency, while maintaining programmability.

Original Document

The original document is available in PDF (uploaded 25 March, 2014 by Andrew Chien).