Abstract
Exact pattern matching is a widely used kernel in many applications. A DRAM-based processing-in-memory (PIM) architecture, Sieve, was recently proposed to alleviate the bottleneck stage of sequence matching in genomics. This paper observes that other exact-pattern-matching-intensive workloads can benefit from a similar architecture. We extend Sieve with several cost-effective modifications, such as a population count logic, chip-level parallelism support, and a hardware data transposition unit, making a general-purpose DRAM-CAM and key-value store that outperforms both CPU and various PIM solutions.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have