How We Reduced Write Latency in TiKV

Explore how PingCAP reduced write latency in TiKV, a distributed transactional Key-Value database, in this informative conference talk. Dive into TiKV's architecture and learn about optimization measures implemented in the raftstore module to lower both average and tail latency. Discover the importance of stable, low-latency write requests in distributed databases and understand the factors that can restrict write performance. Follow along as the speaker explains the use of Asynchronous IO and other optimization techniques to enhance overall latency consistency. Gain insights into the detailed investigation process, the implementation of various optimization measures, and their impact on TiKV's write performance. Examine the results through benchmarks and understand the potential for future improvements in this comprehensive exploration of database optimization strategies.

Introduction
What is TiKV
Why we need low average latency
Why we need low tail latency
Latency of the right phase
Roughstore
Star Series
Source Series
Rough Model
Sequence Diagram
Flowchart
Message Duration
Results
Still duration
Benchmark
Future Work
Outro