Events

Events

[Andy at Sea] Inside the Guts of FloeScan and other Interesting Design Magic

Speaker:
Thomas Kejser
Date:
Mon Oct 19, 2026 @ 04:30pm EDT
Date:
Mon Oct 19, 2026
Time:
04:30pm EDT
Location:
https://cmu.zoom.us/j/92568565952?pwd=N8VKf5xfF5aisG9VoEwdNsCp0dCwel.1Zoom
Title:
Inside the Guts of FloeScan and other Interesting Design Magic
System:
Floe
Video:
YouTube

Talk Info:

The Floe team came from on-premise databases: We wrote our own userspace, collaborative, task-based kernel. We made a database without a buffer pool and created a subsystem to issue I/O in under 1000 cycles/request, outperforming io_uring. Our execution engine maximises joins per second and runs SQL queries with 5MB of query text to completion.

Of course, we needed a query planner too. So, we forked the Postgres planner. After 100K lines of diff and many frustrations, we rewrote it.

Now we are focusing on a new venture, Floe: Making all this work well when the data is on high-latency cloud object storage, with verbose and inefficient lakehouse metadata catalogues.

Floe runs highly complex queries directly on top of Iceberg and Delta lakehouses. In this session, I will deep dive into FloeScan: the component of Floe responsible for indexing cloud storage and reading it at speed and concurrency for users and agents.

This talk is part of the Whatever Andy Found Interesting While Out At Sea.

Bio:

Thomas Kejser is a legendary database engineer and performance optimization expert with over 30 years of experience designing, building, and tuning some of the world's largest database systems. He previously worked at Microsoft, where he helped push SQL Server to its limits and set performance records, and later made significant contributions to the Yellowbrick database engine; he is now working on Floe, bringing his extensive expertise to the next generation of database systems. Armed with an encyclopedic knowledge of database internals and a famously no-nonsense approach to engineering, Thomas is an all-around awesome guy who knows more about making databases go fast than almost anyone on the planet.