

Inside Modern Systems: From Database Performance to AI Infrastructure
Talk 1: Write Combining and Postgres
Melanie will talk about her work introducing write combining into Postgres. She'll cover the benefits of write combining with buffered IO, direct IO, and async IO. She'll walk through the paths that lead to data writes in Postgres, including how these interact with the operating system and kernel. And she'll present benchmarks and data to demonstrate where these processes are bottlenecked in various cases. Her presentation will also cover the differences in write behavior between client backend and various backgrounded processes, such as checkpointer and background writer.
Talk 2: Systems for Future Intelligence: Do Agents Need Abstractions?
While AI agents have revolutionized software engineering, it remains very challenging to build state-of-the-art AI systems. This is in part because of the long tail of workloads in both training and inference, spanning many use cases, models, and hardware. Such diversity demands optimizations tailored to the workload. In this talk, I'll discuss some of our ongoing work on harnessing coding agents to autonomously build reliable and production-quality AI systems.
About Speakers
Melanie is a Postgres hacker and committer working at Microsoft, largely focused on performance work. She has worked on the Postgres executor, planner, storage, statistics, and vacuum subsystems. Most recently she has been hacking on write combining. She also co-organizes PGConf.dev. She is passionate about writing maintainable code and building the Postgres community.
Stephanie Wang joined the faculty of the Paul G. Allen School of Computer Science & Engineering at the University of Washington in fall 2024. She is a co-director of the Systems for Future Intelligence (SyFI) lab, which studies how to build future-proof AI systems. Previously, she received her PhD from UC Berkeley, where she co-created the open-source project Ray, which has been used to train ChatGPT, serve high-performance LLMs, and break the CloudSort 100TB record.
Thank you Fujitsu Enterprise Postgres for sponsoring the event.