Mihail Stoian

Andi Zimmerer | Pruning in Snowflake: Working Smarter, Not Harder Modern cloud-based data analytics systems must efficiently process petabytes of data residing on cloud storage. A key optimization technique in state-of-the-art systems like Snowflake is partition pru...

"The fastest way of processing data is to not process it."

Our SIGMOD 2025 paper shows how Snowflake skips 99.4% of data with new pruning techniques for LIMIT, top-k, and JOIN queries.

Blog: snowflakepruning.github.io
Paper: arxiv.org/abs/2504.11540

@sigmod2025.bsky.social

05.05.2025 05:09 — 👍 5    🔁 1    💬 0    📌 1

SIGMOD BEST PAPER Honorable Mentions
🥇 CRDV: Conflict-free Replicated Data Views
Nuno Faria (INESCTEC & U. Minho)*; José Pereira (U. Minho & INESCTEC)
🥇 DPconv: Super-Polynomially Faster Join Ordering
Mihail Stoian (UTN)*; Andreas Kipf (UTN)

22.04.2025 19:29 — 👍 2    🔁 1    💬 0    📌 0

From top to bottom in the list of accepted papers (2025.sigmod.org/sigmod_paper...):

Yannakakis+
RPT
Galley
LpBound [my favorite]
PDX [my 2nd favorite]
GFTR
Libdbos [my 3rd favorite]
MementoFilter
Spilly

10.04.2025 17:57 — 👍 2    🔁 0    💬 0    📌 0

DPconv just won a SIGMOD'25 Honorable Mention! 🥁

I was quite impressed, given this year's high-quality papers. Let's see who won the big prize.

My list of candidates in the thread below 🧵.

Paper: dl.acm.org/doi/10.1145/...
Slides: stoianmihail.github.io/assets/dpcon...

10.04.2025 17:57 — 👍 3    🔁 0    💬 1    📌 1

🔺Redbench is now live: github.com/utndatasyste....

Let's see how workload-aware your system really is.

09.04.2025 22:11 — 👍 1    🔁 0    💬 0    📌 0

Thrilled to share that we've received the Best Demonstration Award 🏆 at EDBT 2025!

Congratulations to my students @mihailstoian.bsky.social and Ping-Lin Kuo for their excellent work and dedication over the past few weeks—well deserved!

Paper: openproceedings.org/2025/conf/ed...

28.03.2025 13:39 — 👍 4    🔁 1    💬 0    📌 0

Idea: LinDP's limitation lies in its cubic-time DP enumeration strategy, that may enumerate invalid subplans.

Our fix:

1. We output only the valid subplans inspired by DPccp.

2. We also transfer DP-states across linearizations by exploiting how IKKBZ creates them.

13.01.2025 07:20 — 👍 1    🔁 0    💬 0    📌 0

📎 Paper: db.in.tum.de/~birler/pape...

Code: github.com/umbra-db/ada...

Appears at BTW: btw2025.gi.de (the German database conference where "Unnesting Arbitrary Queries" appeared at).

13.01.2025 07:20 — 👍 1    🔁 0    💬 1    📌 0

Umbra's DP optimizer for queries of ~100 relations ran in cubic time.

AWS Redshift's Redset captures a 2,296-relation query.

Our revamped DP enumeration optimizes tree queries like snowflakes of *millions* of relations within 1 sec. 🛸

Joint work w/ Altan Birler & Thomas Neumann.

13.01.2025 07:20 — 👍 1    🔁 0    💬 1    📌 0

If you don't manage to come by, do check out our 3-min presentation:

📹 www.youtube.com/watch?v=B0bU...
📎 openreview.net/forum?id=z7e...

14.12.2024 01:06 — 👍 1    🔁 0    💬 0    📌 0

Are you a fan of Parquet and at #NeurIPS2024 tomorrow? Let's meet at our poster at @trl-research.bsky.social to see how you can reduce your Parquet file sizes by up to 40%.

Virtual compresses tables via functions while ensuring fast column scans.

⏰ 2.30pm
📍East Meeting Room 11 & 12

14.12.2024 01:06 — 👍 5    🔁 3    💬 1    📌 0

Thumbnail: DataLoom: Simplifying Data Loading with LLMs

Vol:17 No:12 → DataLoom: Simplifying Data Loading with LLMs
👥 Authors: Alexander Van Renen, Mihail Stoian, Andreas Kipf
📄 PDF: https://www.vldb.org/pvldb/vol17/p4449-renen.pdf

02.12.2024 05:00 — 👍 3    🔁 2    💬 0    📌 0