Explore a comprehensive analysis of RocksDB key-value workloads at Facebook in this 29-minute conference talk from USENIX FAST '20. Dive into the characterization, modeling, and benchmarking of three typical production use cases: UDB, ZippyDB, and UP2X. Discover interesting findings about key and value size distributions, access patterns, and performance metrics. Learn about the limitations of existing benchmarks like YCSB and understand the proposed key-range based modeling approach for more accurate workload emulation. Gain insights into the challenges of collecting traces in operational environments and the importance of capturing key-space localities for realistic benchmarking of key-value stores.
Characterizing, Modeling, and Benchmarking RocksDB Key-Value Workloads at Facebook
Overview
Syllabus
Introduction
Research Topic
Research Area
Challenges
Methodology
Benchmarking
Production Use Cases
User Profile Use Case
Correct Conversation
Column Families
UDP Column Families
KeySize Variation
KeySize Distribution
CDLCDF
Intensive Needs
Access Count
Use Case
Heat Map
Storage Isles
Data Capture
Results
Heatmap
Key Ranges
Final Results
Conclusion
Use Cases
Weiss SP
Future Improvements
Ads
Replay
Taught by
USENIX
