Workload compression is crucial for database optimization—powering knob tuning, indexing, MV selection, and more. But most prior work targets OLAP with query-only methods. In our new VLDB paper, we present SCompression, a time-sliced workload compression technique for OLTP. The key idea: don’t just choose which queries to keep—preserve concurrency and transaction context so the compressed trace still reflects real system behavior. On a knob-tuning task across real-world and benchmark workloads, SCompression accelerates tuning up to 40× with only ~5% performance impact. Paper: vldb.org/pvldb/vol18/p1865-c… This work was led by visiting PhD student in our group, Baoqing Cai, who will present it at VLDB soon. Happy to discuss—online or in person!

Honored to receive a Google ML & Systems Junior Faculty Award! Grateful for @Google's support as our young lab advances AI- and GPU-driven database systems. This support and recognition will help us turn new ideas into open-source tools and a faster, smarter data infrastructure. Huge thanks to my students and collaborators—their creativity and hard work made this possible. blog.google/products/google-… #GoogleResearch #MLSystems #GPUDatabases #AI

Thrilled to share that our project "Accelerating Large-Scale Data Analytics with Multi-GPU Resource Redistribution" in collaboration with @TalatiNishil, has been selected for the @nvidia Academic Grant Program (nvidia.com/en-us/industries/…)! 🚀 Shout-out to @NVIDIAAIDev for their generous allocation of A100 GPU hours! These resources will supercharge our research on GPU-accelerated databases and help us push the boundaries of large-scale data analytics. #NVIDIAGrant #GPU #A100 #DataAnalytics #GPUDatabases

GPU memory capacity and I/O bandwidth limitations have been major bottlenecks for GPU-accelerated databases. We tackle such a challenge in our latest @VLDBconf 2025 paper: vldb.org/pvldb/vol18/p1250-y… We present a novel resource-sharing method for multi-GPU systems that enables GPU acceleration on datasets that far exceed GPU memory capacity, without relying on GPU-side data caching. The main idea is to repurpose underutilized IO resources from AI workloads to accelerate data movement for data analytics. Our approach delivers over 2x better price-performance than CPU-based solutions. This work is led by Yichao Yuan (@UMichCSE PhD student) in collaboration with @TalatiNishil. We will be presenting this at VLDB 2025 in London later this summer. Hope to see you there!

I'm recruiting PhD students for my lab @UMichCSE on applied ML for databases, starting Fall 2023. My research and industry experience convinces me that database automation and simplification are more important than ever, and ML/AI can play a big role. Please apply and reach out!

Just to add that, as part of this transition, I recently joined @databricks to work on the Delta Lake/Lakehouse. I've moved to Ann Arbor, but will visit the DB SF/SV office periodically. Look forward to hanging out and collaborating with friends both in A2 and the bay area:)

Personal update: I'll be joining the University of Michigan, Ann Arbor in the Division of Computer Science and Engineering @UMichCSE as an Assistant Professor in Fall 2023! Thrilled to work on exciting database research with everyone at @UMich!

.@CMUDB Intro to Database Systems 2021 ➜ Lecture #23: Distributed OLAP Database Systems Video: youtube.com/watch?v=t6dpjpQx… Slides: 15445.courses.cs.cmu.edu/fal… Execution models, query planning, and distributed join algorithms.

.@CMUDB Intro to Database Systems 2021 ➜ Lecture #22: Distributed OLTP Database Systems Video: youtube.com/watch?v=n4TEdXZ8… Slides: 15445.courses.cs.cmu.edu/fal… Commit protocol, replication, and CAP theorem.

.@CMUDB Intro to Database Systems 2021 ➜ Lecture #21: Introduction to Distributed Databases Video: youtube.com/watch?v=h838F6i8… Slides: 15445.courses.cs.cmu.edu/fal… Architecture and partitioning.

.@CMUDB Intro to Database Systems 2021 ➜ Lecture #20: Crash Recovery Algorithms Video: youtube.com/watch?v=0wocj5Tp… Slides: 15445.courses.cs.cmu.edu/fal… Checkpointing and ARIES.

.@CMUDB Intro to Database Systems 2021 ➜ Lecture #19: Logging Protocols + Schemes Video: youtube.com/watch?v=mD5znxu4… Slides: 15445.courses.cs.cmu.edu/fal… Shadow paging and write-ahead logging.

.@CMUDB Intro to Database Systems 2021 ➜ Lecture #18: Multi-Version Concurrency Control Video: youtube.com/watch?v=QXWf1sgw… Slides: 15445.courses.cs.cmu.edu/fal… With demo on Postgres under different isolation levels!

.@CMUDB Intro to Database Systems 2021 ➜ Lecture #17: Timestamp Ordering Concurrency Control Video: youtube.com/watch?v=6UbWb9YE… Slides: 15445.courses.cs.cmu.edu/fal… Basic timestamp ordering protocol and optimistic concurrency control.

.@CMUDB Intro to Database Systems 2021 ➜ Lecture #16: Two-Phase Locking Concurrency Control Video: youtube.com/watch?v=4O1-tXw4… Slides: 15445.courses.cs.cmu.edu/fal… With deadlock detection demos on Postgres and MySQL.

.@CMUDB Intro to Database Systems 2021 ➜ Lecture #15: Concurrency Control Theory Video: youtube.com/watch?v=7u-XH_2g… Slides: 15445.courses.cs.cmu.edu/fal… Transactions and serializability.