Speakers:

Benjamin Owad , Jiaying Li , Bowei Chen

Date:

Tue Oct 20, 2026 @ 12:00pm EDT

Date:

Tue Oct 20, 2026

Time:

12:00pm EDT

Location:

NSH 4305

Title:

Inside Snowflake's Query Engine: Optimizing Scans, Semi-Structured Data, and DML

System:

Snowflake

---

Talk Info:

How data is read, written, and stored largely determines the performance of an analytical query engine. Snowflake in particular favors adaptivity over user-facing knobs, so the engine is responsible for making many decisions at runtime. This session presents recent work in these areas, beginning with table scans. We discuss an adaptive scan parallelism technique, pruning strategies, I/O request coalescing, workload-based column placement, and new work on in-memory workloads. Next, we describe our work on semi-structured VARIANT data and subcolumnarization, including path-level Bloom filters and future work in workload-driven adaptivity. Finally, we cover optimizations for updates, deletes, and merges on immutable files, along with skew handling techniques and aggregation optimizations in distributed execution. Together, these efforts improve how Snowflake stores, accesses, and modifies data.

Bio:

Benjamin Owad is a computer science enthusiast based in the San Francisco Bay Area working at Snowflake. He studied computer science at Carnegie Mellon University, concentrating in systems. His interests span databases, open-source software, vintage hardware, and niche operating systems

Jiaying Li is a software engineer at Snowflake. Prior to that she was an MCDS (Master of Science in Computational Data Science) student at Carnegie Mellon University School of Computer Science.

Bowei Chen is a software engineer at Snowflake Computing. He was a master's student from the vaunted Carnegie Mellon Database Group in 2017. Before that, he was (and possibly still remains) the best student to graduate from Shanghai Jiao Tong University.