An Energy-Efficient Domain-Specific Architecture for Regular Expressions

Abstract

Regular Expressions (REs) are a computational kernel widely used for finding patterns in data in compute-intensive tasks such as genomic markers research, signature-based detection, and database query. Although flexible on the set of searched REs, software-based solutions cannot fulfill latency or throughput requirements to analyze massive data volumes at a given power budget. For this reason, many approaches exploit hardware accelerators as an offloading engine for REs matching. Indeed, various solutions rely on FPGA reconfigurability to embed automata into the reconfigurable fabric. However, this approach leads to time-consuming updates of the REs to search. This work exploits REs as sequences of basic instructions and builds a Domain-Specific Architecture (DSA), called TiReX, for RE matching on FPGAs. Our approach enables the user to change the desired RE at run-time, providing software programmability, flexibility, and specialized hardware mechanisms. Our DSA delivers performance in line with other state-of-the-art hardware approaches, while providing remarkable flexibility and we underline the importance of energy efficiency for these computations. We compared with multiple state-of-the-art software obtaining remarkable performance while achieving noticeable results with a better energy efficiency that ranges from 3× to 490× with our multi-core.

YARB: a Methodology to Characterize Regular Expression Matching on Heterogeneous Systems

“…On the one hand, some solutions embed the automata logic directly on the reconfigurable fabric exploiting the adaptability at the cost of regenerating new bitstreams for unseen REs [16], [18], [37], [38]. On the other hand, others devise programmable DSAs to exploit the domain specialization offered by the reconfigurable hardware and expose flexibility in terms of instructions/REs to execute [22], [39], [40]. Despite some solutions are open-source, they are not documented, cumbersome to use, or do not provide means to really execute on hardware; thus, we exclude them from our work.…”

Enabling Efficient Regular Expression Matching at the Edge through Domain-Specific Architectures

“…We analyze the State of the Art of REs matching hardware solutions. On the one hand, many embed the REs/automata either in the reconfigurable fabric or in special memories [16], [17], while others exploit SW-programmable DSAs [14], [18]. Given the limited resources we exclude embedding solutions; therefore, we build our system design approach for HIDS on the open DSA CICERO [14] that features flexibility for pattern detection and specialization for low latency.…”

“…CICERO builds on the fundamental approach of seeing the REs as a sequence of operations executed on a sequence of characters [14], [18], [19], where it specifically execute parallel threads that working in lockstep resemble a breadthfirst exploration. Moving to the kernel of CICERO, the engine is the basic unit that can perform the RE matching on an input string and returns whether it is accepted or rejected.…”

“…Exploiting IoT and CPS systems for information and data retrieval is essential for knowledge extraction [1], and REs represent an excellent mean for pattern analysis [7]. Several approaches exist to perform RE matching efficiently (or its corresponding automata-based version) [16], [18], [24], [25]. However, the majority of the solutions target highend, datacenter-like application scenarios, lacking an effective solution to constrained deployments like our case.…”