SEPH: Scalable, Efficient, and Predictable Hashing on Persistent Memory
summary
Implements a new hash table for persistent memory that combines a novel Level Segment (LS) layout, two low‑overhead splits including a one-third split, and semi lock-free concurrency to cut PM writes and resizing work. Evaluated on real PM hardware, it claims up to 15.4× higher throughput, scales to 48 threads, and reduces tail latency by up to 19.3×. Microbenchmarks show the semi‑lock‑free control and reference‑free rehashing together halve resizing cost and substantially lower PM write amplification.
AI-written summary · about
chapters (8)
- 0:00Introduction to PM technology
- 1:20Existing PM hashing schemes
- 3:20Performance dilemmas
- 5:32The Seth approach
- 7:06Low-overhead techniques
- 9:29Concurrency control
- 10:27Experimental evaluation
- 12:44Conclusion
related talks
24mWrite-Optimized Dynamic Hashing for Persistent Memory
20mRevisiting Secondary Indexing in LSM-based Storage Systems with Persistent Memory
01h 05mImplementing Understandable World Class Hash Tables in C++
43mPromise and Pitfalls of Persistent Memory
35mDeveloping Software for Persistent Memory
35mDeveloping Software for Persistent Memory
40m"Gaining Constant time Lookup over Unorganized Data"
29mSLM-DB: Single-Level Key-Value Store with Persistent Memory