Title: How TiDB Keeps Performance Predictable under a High Load
Time: August 25, 6 PM Pacific Daylight Time. Doors will open at 5:30 PM.
Join our Zoom Webinar: https://pingcap.zoom.us/j/92959043179
Introduction:
Predictable performance at any scale is important for mission critical applications. TiDB is a distributed SQL database that provides excellent horizontal scalability. As shown in the 2022 VLDB paper “TAOBench: An End-to-End Benchmark for Social Networking Workloads,” TiDB achieves fairly good throughput and low latency under high workloads.
In this talk, Jinpeng will reveal the secrets behind TiDB’s predictable performance under high load. This includes load-based splitting and scheduling to evenly distribute load between nodes, write traffic control with feedback to prevent overwhelming the underlying engine, and throttling background jobs to improve tail latency.
Speaker
Jinpeng Zhang
Head of the TiDB storage team, PingCAP
Jinpeng Zhang joined PingCAP in 2016. His work focuses on storage engine and performance optimization. He has made many improvements to the performance of TiDB, including updates to the Raft Engine and write scheduler. He also made many optimizations to TiDB’s transaction mechanism, such as pushing down the insertion of existing checks, embedded short values, and designing fast resolving locks. He is a maintainer of TiKV and TiDB, and a contributor to RocksDB.