Maximilian Kuschewski — optimizing performance at TigerBeetle

DESCRIPTION

PUBLICATIONS

1. VLDB 2026 · Research paper

   BtrLog: Low-Latency Logging for Cloud Database Systems

   Maximilian Kuschewski, Lam-Duy Nguyen, Matthias Jasny, Tobias Ziegler, Viktor Leis, Muhammad El-Hindi

   pdf code

   +−abstract

   Cloud database systems cannot rely on instance-local disks for WAL durability, forcing the log onto remote storage. Block storage is easy to adopt but slow and expensive; object storage is cheap but impractical for OLTP. BtrLog is a reusable cloud logging service for the common single-writer architecture: it replicates log records across a quorum of SSD-backed nodes in one network round trip, then archives large segments to object storage off the latency-critical path. It beats EBS on latency and lifts end-to-end transaction throughput when wired into a DBMS.

2. CIDR 2026 · Demo paper

   Cloudspecs: Cloud Hardware Evolution Through the Looking Glass

   Till Steinert, Maximilian Kuschewski, Viktor Leis

   pdf code demo

   +−abstract

   A decade of cloud hardware, 2015–2025: network bandwidth improved per dollar by an order of magnitude, CPU and DRAM much less, NVMe performance stagnant since 2016. Smaller providers undercut the big three by up to 5× on commodity VMs. The paper maps shifting bottlenecks for cost-efficient data-intensive systems; an interactive tool explores the instance landscape.

3. VLDB 2025 · Research paper · Best Paper Award

   AnyBlox: A Framework for Self-Decoding Datasets

   Mateusz Gienieczko, Maximilian Kuschewski, Thomas Neumann, Viktor Leis, Jana Giceva

   pdf code

   +−abstract

   New storage encodings keep winning on paper and losing in production: implementation cost and compatibility freeze the formats we actually use. AnyBlox bundles lightweight WebAssembly decoders with the data, decoupling decoders from both systems and file-format specs. Formats can evolve, encodings can be instance-optimized, and research can ship. It plugs into DuckDB, Spark, and Umbra without giving up performance or safety.

4. SIGMOD 2025 · Research paper

   High-Performance Query Processing with NVMe Arrays: Spilling without Killing Performance

   Maximilian Kuschewski, Jana Giceva, Thomas Neumann, Viktor Leis

   pdf

   +−abstract

   In-memory engines are fast until they aren't; engines that spill to disk are robust and slow. Adaptive materialization turns any hash-based in-memory operator into an out-of-memory operator without punishing the in-memory path. Self-regulating compression tunes spilling throughput to the current workload and hardware. Spilly, the prototype, matches state-of-the-art in-memory systems and still executes large out-of-memory workloads by spilling to NVMe arrays.

5. SIGMOD 2023 · Research paper

   BtrBlocks: Efficient Columnar Compression for Data Lakes

   Maximilian Kuschewski, David Sauerwein, Adnan Alhomssi, Viktor Leis

   pdf code

   +−abstract

   Open formats like Parquet were not designed for remotely accessed data lakes or today's networks. Decompression becomes CPU-bound; scans get slower and more expensive. BtrBlocks is an open columnar format built on lightweight encodings: fast decompression, high compression ratios, meant for data lakes on blob storage.

6. VLDB 2021 · Vision paper

   Towards Cost-Optimal Query Processing in the Cloud

   Viktor Leis, Maximilian Kuschewski

   pdf code

   +−abstract

   Public clouds offer hundreds of heterogeneous instances; for analytics, the wrong shape can cost orders of magnitude more. A simple model of workload, hardware, and price picks the optimal configuration — and should also steer how cloud-native database systems themselves evolve.

7. CIDR 2021 · Short paper

   White-Box OLAP Performance Modeling for the Cloud

   Maximilian Kuschewski, Viktor Leis

   pdf

   +−abstract

   Companion to the cost-optimal vision: a white-box model of workload, hardware, and cost that selects instance configurations for cloud OLAP, and an argument that this kind of model should guide the next generation of cloud-native engines.

EDUCATION

WORKING EXPERIENCE

SEE ALSO

• github.com/maxi-k
• scholar.google.com
• linkedin.com/in/maxi-k
• x.com/maxikuschewski
• tigerbeetle.com

COLOPHON

Static HTML and CSS. Set in JetBrains Mono. Press t to toggle color scheme. Theme preference lives in localStorage. No cookies.