October Edition: Postgres and Tigerbeetle

We're excited to be back with another evening of Database Internals! Expect great talks, pizza 🍕 and plenty of time to chat.

📍 CHECK24, Erika-Mann-Straße 62-66
📅 Tuesday, October 27, doors open at 18:30

Databricks: PostgreSQL as an Open Data Format: 100x Faster TPC-H Queries Through Direct Storage Reads
PostgreSQL has never been known as an analytical system. Earlier efforts forked it to add columnar storage, MPP, and vectorized execution, drifting away from core PostgreSQL and breaking compatibility with its tooling. More recent extensions federate to engines like DuckDB, but still hit the limits of an executor designed for OLTP.
This talk flips the model: rather than going through the PostgreSQL execution layer, read the data directly from storage. Borrowing from lakehouse architectures (open file formats like Parquet, table formats like Iceberg, governance through a catalog), Hristo Stoyanov, Member of the Technical Staff at Databricks, introduces the "lakebase": separating storage and compute for operational databases so that multiple engines can perform serverless reads of PostgreSQL's own file format. We'll start with the state of analytics on PostgreSQL today, review how lakehouse engines use open formats and catalogs to stay fast while respecting transactional and access control guarantees, then dig into the internals of a new open source project that reads PostgreSQL data files outside the PostgreSQL process. The talk closes with a demo showing up to 100x faster TPC-H on PostgreSQL data, with no impact on the running instances.
Hristo Stoyanov works on the Lakebase team at Databricks. Previously Google Spanner, SingleStore and Stripe.

TigerBeetle: The Tale of Taming TigerBeetle's Tail Latency
Learn how we reduced TigerBeetle's tail latency through algorithm engineering. 'Algorithm engineering goes beyond studying theoretical complexity and considers how algorithms are executed efficiently on modern super-scalar CPUs. Specifically, we will look at Radix Sort and a k-way merge and explore how to implement them efficiently. We then demonstrate how we apply these algorithms incrementally to avoid latency spikes in practices.
Tobias Ziegler is a Software Engineer at TigerBeetle, with a Ph.D. from TU Darmstadt and a postdoc at TU Munich.

Agenda
18:30 Doors open
18:40 Welcome
18:45 - 19:10 Talk + Q&A (CHECK24)
19:10 - 19:55 Talk + Q&A (TigerBeetle)
19:55 - 20:15 Pizza break 🍕
20:15 - 21:00 Talk + Q&A (Databricks)
21:00 Wrap-up & networking