POLARDB Meets Computational Storage: Efficiently Support Analytical Workloads in Cloud
summary
Paper reports a deployed replacement of SSDs with FPGA-based computational storage drive devices inside Alibaba Cloud POLARDB, plus coordinated changes to the analytics engine, filesystem and device driver. Devices implement transparent compression (Snappy) in hardware, push table-scan logic into drives, add a lightweight header to block format, and use batching to raise FPGA utilization. Evaluation on a 7‑DB/3‑storage node cluster shows random IOPS >2× vs a leading SSD, CPU use drops, 30% latency reduction for 11 queries and >20% PCIe traffic cut for 7 queries.
AI-written summary · about
chapters (11)
- 0:00Intro
- 0:31POLARDB: Cloud-Native Relational Database
- 2:51Table Scan in Storage Nodes
- 5:47Computational Storage: A Very Simple Idea
- 9:10CSD with Data Path Transparent Compression
- 11:20POLARDB Meets Computational Storage
- 15:20Computational Storage Drive Implementation
- 16:54Implementation & Evaluation
- 19:12System Evaluation
- 20:27Conclusion
- 33:02Unit Test on Storage Node
related talks
21mPolarDB: Alibaba Cloud-Native Relational Database Service
12mAdvancing Flash Storage
19mThree Actionable Ways to Improve your Cloud Storage
20mBuilding An Elastic Query Engine on Disaggregated Storage
16mSolidigm CSAL Solution Brings Advanced IO Shaping, Caching and Data Placement Into NVIDIA DPU DOCA S
28mIntroducing AlloyDB, a PostgreSQL-compatible cloud database service
06mUsing FPGA's and Xeon processors to accelerate cloud workloads
24mCloud Workloads At Petabyte Scale