Systems Research Mini Workshop (Reception starting at 1:30pm Siebel 2407)
2:00pm, Siebel 2405
Transparently Enhancing Applications through Operating System Architectures and Mechanisms (Jing Liu)
Abstract
Operating systems are the foundational software that all modern applications rely on, bridging applications and hardware. As both applications and hardware have rapidly evolved over the past decades, it has become imperative for operating systems to identify and support new demands without breaking existing applications. My research focuses on the progressive evolution of operating systems to achieve high performance, stronger reliability, and better customization, with an emphasis on new architectures and mechanisms.
In this talk, I will present our work on evolving operating system architectures for high performance, where we build a filesystem, uFS, as a semi-microkernel filesystem running in userspace for performance while coexisting with commodity OSes. I will then introduce Ananke, a new recovery mechanism that enables transparent runtime recovery from filesystem failures, providing stronger reliability guarantees. Finally, I will present our recent work, Midas, which introduces a systematic tunability management framework that enables dynamic runtime tuning of previously untunable logic – without kernel modification or application disruption – while supporting rich customization semantics.
Speaker bio
Jing Liu is a Senior Researcher at Microsoft Research Asia. She received her Ph.D. in 2024 from the University of Wisconsin–Madison, advised by Prof. Andrea Arpaci-Dusseau and Prof. Remzi Arpaci-Dusseau. Jing’s research focuses on evolving operating systems, as well as file and storage systems, to meet modern demands. She has received the Erik Riedel Best Paper Award at FAST 2025 and a Meta PhD Research Fellowship in 2022.
3:00pm, Siebel 2405
LiquidCache: Efficient Pushdown Caching for Cloud-Native Data Analytics (Xiangpeng Hao)
Abstract
We present LiquidCache, a novel pushdown-based disaggregated caching system that evaluates filters on cache servers before transmitting data to compute nodes. Our key observation is that data decoding, not filter evaluation, is the primary bottleneck in existing systems. To address this challenge, we transcode Parquet data into a lightweight “Liquid” format optimized for caching and filter evaluation. This format is co-designed with filter evaluation semantics to enable selective decoding, late filter materialization, and encoding-aware filter evaluation, delivering low decoding costs while preserving high compression ratios. The “Liquid” format exists exclusively in the cache, allowing easy adoption without breaking ecosystem compatibility. Through integration with Apache DataFusion and evaluation with ClickBench and TPC-H, we demonstrate that LiquidCache reduces cache CPU time by up to 10× without increasing memory footprint, and reduces network traffic by two orders of magnitudes compared to non-pushdown systems.
Speaker bio
Xiangpeng Hao (https://xiangpeng.systems/) is a fifth-year PhD student at UW-Madison advised by Andrea Arpaci-Dusseau and Remzi H. Arpaci-Dusseau. His research focuses on LiquidCache (https://github.com/XiangpengHao/liquid-cache), a distributed caching system for cloud-native analytical workloads. LiquidCache has been actively tested in production by organizations such as Pydantic, OpenObserve, and AWS internal systems. He has also self-raised research funding from InfluxData, SpiralDB, and Bauplan to support his PhD work.
4:00pm, Siebel 2405
Fray: An Efficient General-Purpose Concurrency Testing Platform for the JVM (Ao Li)
Abstract
Concurrency bugs are hard to discover and reproduce, even in well-synchronized programs that are free of data races. Thankfully, prior work on controlled concurrency testing (CCT) has developed sophisticated algorithms to effectively search over the space of thread interleavings. Unfortunately, in practice, these techniques cannot easily be applied to real-world Java programs due to the difficulties of controlling concurrency in the presence of the managed runtime and complex synchronization primitives. So, mature Java projects that make heavy use of concurrency still rely on naive repeated stress testing in a loop. This talk will present Fray, our new framework for push-button concurrency testing of JVM programs that is explicitly designed with the goal of general-purpose applicability. Fray employs a concurrency control mechanism we call “shadow locking”, which guarantees soundness and completeness of expressiveness under a reasonable set of assumptions, and is more efficient than classical OS-level or JVM-simulation-based approaches to managing concurrency. Fray works out-of-the-box on production-grade software such as Apache Kafka, Apache Lucene, and Google Guava—we have shown how hundreds of existing unit tests in these mature projects can fail under certain interleavings, and have helped developers confirm and debug over a dozen new bugs.
Speaker bio
Ao Li is a PhD candidate in the School of Computer Science at Carnegie Mellon University. He works on designing and building tools that make complex systems easier to debug and test. He has helped uncover hundreds of software bugs in widely-used open-source and commercial software, and his tools are being used by companies like Antithesis, Amazon, and Microsoft.