Explore a conference talk that introduces SplinterDB, a novel key-value store designed specifically for NVMe solid state drives. Learn about the STBε-tree data structure, which improves I/O and CPU concurrency while reducing write amplification without compromising query performance. Discover how SplinterDB combines elements from log-structured merge trees and Bε-trees to optimize compaction and minimize CPU costs. Examine the highly concurrent memtable and cache design that reduces cache misses. Compare SplinterDB's performance to RocksDB, a leading key-value store, and understand how it achieves 6-10x faster insertions, 2-2.6x quicker point queries, and a 2x reduction in write amplification. Gain insights into the technical aspects of SplinterDB's architecture and its potential impact on modern storage systems.
SplinterDB - Closing the Bandwidth Gap for NVMe Key-Value Stores
Overview
Syllabus
Introduction
Overview
Performance
Size 2Bepsilon tree
Size 3Bepsilon tree
Size 4Bepsilon tree
Flush Compact
Conclusion
Taught by
USENIX
Related Courses
-
AC-Key - Adaptive Caching for LSM-based Key-Value Stores
-
MatrixKV - Reducing Write Stalls and Write Amplification in LSM-tree Based KV Stores
-
Improving Performance of Flash Based Key-Value Stores Using Storage Class Memory as a Volatile Memory Extension
-
RubbleDB - CPU-Efficient Replication with NVMe-oF
-
PinK - High-speed In-storage Key-value Store with Bounded Tails
-
SPEICHER - Securing LSM-based Key-Value Stores Using Shielded Execution
