R&D: H-Rocks, CPU-GPU accelerated Heterogeneous RocksDB on Persistent Memory

ACM Digital Library has published, in Proceedings of the ACM on Management of Data, an article written by Shweta Pandey, and Arkaprava Basu, Indian Institute Of Science, Bengaluru, Karnataka, India.

Abstract: “Persistent key-value stores (pKVS) such as RocksDB are critical to many internet-scale services. Recent works leveraged persistent memory (PM) to improve pKVS throughput. However, they are typically limited to CPUs.“

“We develop H-Rocks to judiciously leverage both the CPU and the Graphics Processing Unit (GPU) for accelerating a wide range of RocksDB operations. H-Rocks selectively accelerates performance-critical parts of RocksDB on the GPU. It uses operation sub-batching and key-value versioning to leverage GPU’s parallelism while maintaining compatibility with RocksDB. It harnesses GPU’s high-bandwidth memory while limiting data movement between the CPU and GPU. In YCSB workloads, H-Rocks outperforms CPU-based pKVSs like Viper, Plush, and pmem-RocksDB by 3-18×.