Explore the innovative HotRing key-value store designed to address hotspot issues in in-memory systems. Delve into the challenges of hotspot-awareness in existing key-value stores and discover how HotRing optimizes performance for highly skewed workloads. Learn about the ordered-ring hash index structure, which improves access to hot items, and the lightweight strategy for detecting hotspot shifts at runtime. Understand the comprehensive lock-free design that enables HotRing to leverage multi-core architectures for massively concurrent requests. Examine the experimental results showcasing HotRing's significant performance improvements compared to other in-memory key-value stores on skewed workloads. This conference talk, presented at FAST '20 by Le Cai from Alibaba Group, offers valuable insights into cutting-edge solutions for enhancing the efficiency and reliability of in-memory key-value stores in the face of increasingly serious hotspot issues.
Overview
Syllabus
Introduction
Problem Statement
Solutions
Hash Index
Algorithm
Problems
Strategy
Rehashing
Evaluation
Comparisons
Performance
Stability
Latency
Conclusion
Taught by
USENIX
Related Courses
-
ROLEX - A Scalable RDMA-oriented Learned Key-Value Store for Disaggregated Memory Systems
-
Fast RDMA-based Ordered Key-Value Store Using Remote Learned Cache
-
Characterizing, Modeling, and Benchmarking RocksDB Key-Value Workloads at Facebook
-
Improving Performance of Flash Based Key-Value Stores Using Storage Class Memory as a Volatile Memory Extension
-
ScaleXFS - Getting Scalability of XFS Back on the Ring
