Lessons Learned from benchmarking and profiling distributed PostgreSQL | POSETTE 2024

From the creator

Video of a conference talk about lessons learned from benchmarking and profilling distributed PostgreSQL presented by Lotte Felius at POSETTE: An Event for Postgres 2024. In this talk, I’ll explain how I investigated the performance of Citus using the Citus Benchmark Suite and YCSB. Quantifying performance however is not straightforward since Citus is a distributed system where queries are distributed across multiple worker nodes. Therefore, I’ll explain some of my learnings as a former intern with no prior experience in benchmarking Databases, which includes: - The pros and cons of different architectures used for distributed database systems based on PostgreSQL - Why quantifying performance is harder for distributed database systems and how you can use benchmarks to get an accurate indication of the performance - How to identify bottlenecks based on benchmarking results and which optimizations you can possibly to do mitigate them Lotte is currently a second year PhD student at the “Centrum voor Wiskunde & Informatica” (CWI), based in Amsterdam, Netherlands. She is in the Database Architectures group and her current research includes secure query processing, lightweight compression on GPUs and secure next-gen file formats. Former intern at Citus (Microsoft). Chapters: 00:00 - Intro 01:44 - Architectures for Distributed PostreSQL 05:53 - The PostgreSQL Query Processing Pipeline 08:56 - The scope of the research 11:09 - Benchmarking Citus 17:41 - Identifying bottlenecks 22:

Published 2024-06-13 · Watch on YouTube ↗
Availability and playback restrictions are controlled by the original channel.