Oana Balmau
Oana Balmau is a computer systems researcher whose work on storage and data management supports machine learning workloads. After bachelor's and master's degrees in computer science at EPFL, she completed a PhD at the University of Sydney under Willy Zwaenepoel; her SILK key-value store won Best Paper at USENIX ATC in 2019, and her dissertation received the 2021 CORE John Makepeace Bennett Award and an honorable mention for the SIGOPS Ritchie Doctoral Dissertation Award. She joined McGill University in January 2021, co-founded the MLPerf Storage benchmark, and co-authored Falcon, a Best Paper winner at the 2024 ACM/IEEE Symposium on Edge Computing. Recent papers such as MinatoLoader and HarMoEny target faster ML training and MoE model inference. She is an assistant professor at McGill, where she leads the DISCS Lab, and an associate academic member of Mila.
- Now at
- School of Computer Science, McGill University, Mila and University of Toronto
Timeline
-
By 2019
EPFL degrees and a Sydney PhD
She earns bachelor's and master's degrees in computer science at EPFL in Switzerland. By 2019, she is at the University of Sydney, where Willy Zwaenepoel advises her PhD.
-
2019
SILK wins best paper
In 2019, she is first author of SILK, a RocksDB-derived key-value store with an I/O scheduler that reaches up to two orders of magnitude lower 99th percentile latencies than RocksDB and TRIAD. The paper, written with Zwaenepoel and colleagues from Nutanix and IBM Research–Zurich, wins Best Paper at the USENIX Annual Technical Conference.
-
January 2021
Assistant professor at McGill
She joins McGill's School of Computer Science in January 2021 as an assistant professor.
-
2021
Two dissertation honours
In 2021, her Sydney dissertation, Redesigning Persistent Key-Value Stores for Future Workloads, Hardware, and Performance Requirements, wins the CORE John Makepeace Bennett Award for the best computer science dissertation in Australia and New Zealand. It also earns an honorable mention for the ACM SIGOPS Dennis M. Ritchie Doctoral Dissertation Award.
-
2023
Co-founding MLPerf Storage
Within MLCommons, she co-founds MLPerf Storage, an open-source benchmark for storage on ML workloads. As Storage Working Group co-chair, she reports in September 2023 that the first round has over 28 performance results from five companies.
-
2024
Falcon wins best paper
She co-authors Falcon: Live Reconfiguration for Stateful Stream Processing on the Edge, which wins the Best Paper Award at the ninth ACM/IEEE Symposium on Edge Computing, held in Rome in December 2024.
-
2025–2026
Data loading and MoE inference
In 2025, she co-authors MinatoLoader, a PyTorch data loader that builds batches by prioritizing fast-to-preprocess samples so GPUs sit idle less. In 2026, she co-authors HarMoEny, on efficient inference of MoE models, which appears at the IEEE International Parallel and Distributed Processing Symposium.
-
As of October 2026
Leading McGill's DISCS Lab
She is an assistant professor in McGill's School of Computer Science, where she leads the DISCS Lab, and an associate academic member of Mila. She also holds a status-only appointment in computer science at the University of Toronto, and her research focuses on storage systems and data management for ML, data science, and edge computing workloads.
Career
- Since 2021McGill UniversityAssistant professor
- NowSchool of Computer Science, McGill UniversityLead
- NowMilaAssociate academic member
- NowUniversity of TorontoStatus-only appointment in computer science
- From 2023MLCommonsStorage Working Group co-chair
Education
- EPFLBachelor's degree in computer science
- University of SydneyPhD
Sources and credits
Photo: We have no published photo of Oana Balmau yet. Send us a free-licensed photo.
- 2021 Invited Speakers
- SILK: Preventing Latency Spikes in Log-Structured Merge Key-Value Stores | USENIX
- Oana Balmau | Mila
- Prof. Oana Balmau receives SIGOPS distinction
- The Dennis M. Ritchie Thesis Award 2021
- 5 minutes with Oana Balmau
- MLCommons publishes storage benchmark for AI
- SEC 2024 | 9th ACM/ IEEE Symposium on Edge Computing
- MinatoLoader: Accelerating Machine Learning Training Through Efficient Data Preprocessing

