Vincent Lannurien
d822cdaa6a
5.3 KiB
title, venue, year, date, tags, paper_count, draft
| title | venue | year | date | tags | paper_count | draft |
|---|---|---|---|---|---|---|
| TOCS 2025 Digest | TOCS | 2025 | 2025-01-01 | 10 | false |
10 papers selected.
Whole-system Persistence Made Efficient with Tree-structured Checkpointing on Microkernel
Mingkai Dong, Fangnuo Wu, Gequan Mo, Haibo Chen
TL;DR — A microkernel-based whole-system persistence scheme uses tree-structured incremental checkpointing to achieve low-overhead, crash-consistent snapshots of the entire OS state.
Why notable — Whole-system persistence is a foundational building block for reliable systems; this paper shows it can be done efficiently within a microkernel architecture.
XpuTEE: A High-Performance and Practical Heterogeneous Trusted Execution Environment for GPUs
Shulin Fan, Zhichao Hua, Yubin Xia, Haibo Chen
TL;DR — XpuTEE extends trusted execution environments to GPUs by designing a hardware-assisted isolation mechanism that protects GPU computations with low performance overhead.
Why notable — As GPUs process sensitive ML workloads in shared clouds, TEE support for accelerators is urgently needed; XpuTEE is a comprehensive and practical solution.
RegVault II: Achieving Hardware-Assisted Selective Kernel Data Randomization for Multiple Architectures
Ruorong Guo, Yangye Zhou, Jinyan Xu, Wenbo Shen et al.
TL;DR — RegVault II uses hardware features to selectively randomize sensitive kernel data structures at runtime across multiple ISAs, raising the bar for kernel exploitation.
Why notable — Kernel data-only attacks bypass existing code-randomization defenses; this work's multi-architecture approach makes selective data randomization practical for production kernels.
Validating JIT Compilers via Compilation Space Exploration
Cong Li, Yanyan Jiang, Chang Xu, Zhendong Su
TL;DR — Compilation space exploration systematically generates and tests the large space of valid JIT compilation outcomes to find miscompilation bugs in production JIT compilers.
Why notable — JIT correctness is notoriously hard to test; this paper's systematic exploration strategy finds real bugs in widely-used runtimes and advances the state of compiler validation.
Freezing-based Memory and Process Co-design for User Experience on Resource-limited Mobile Devices
Changlong Li, Zongwei Zhu, Chun Jason Xue, Yu Liang et al.
TL;DR — A co-designed memory and process management scheme freezes background processes at fine granularity to reclaim memory while preserving fast resume latency on constrained mobile hardware.
Why notable — Mobile memory pressure directly degrades user experience; this paper's co-design perspective yields measurable improvements on real devices with limited resources.
Analyzing Configuration Dependencies of File Systems
Tabassum Mahmud, Om Rameshwar Gatla, Duo Zhang, Carson Love et al.
TL;DR — This work systematically analyzes the dependency graph among file-system configuration options to reveal hidden interactions that lead to silent data corruption or crashes.
Why notable — File-system misconfiguration is a major source of data loss in practice; understanding configuration dependencies is essential for building safer storage systems.
Efficient Fault Tolerance for Stateful Serverless Computing with Asymmetric Logging
Sheng Qi, Haoyu Feng, Xuanzhe Liu, Xin Jin
TL;DR — Asymmetric logging decouples the logging cost between the fast and slow paths of stateful serverless functions, enabling low-overhead fault tolerance without sacrificing recovery guarantees.
Why notable — Fault tolerance for stateful serverless remains an open performance challenge; this paper's asymmetric design significantly reduces logging overhead compared to symmetric approaches.
Towards Serialization/Deserialization-free State Transfer in Serverless Workflows
Xingda Wei, Fangming Lu, Zhuobin Huang, Rong Chen et al.
TL;DR — This system eliminates serialization and deserialization costs when passing state between serverless functions by enabling direct in-memory state transfer across workflow stages.
Why notable — Ser/deser overhead is a dominant cost in serverless workflows; removing it fundamentally changes the performance profile of function chaining at scale.
Enabling Anonymous Online Streaming Analytics at the Network Edge
Yunming Xiao, Yanqi Gu, Yibo Zhao, Sen Lin et al.
TL;DR — This paper designs an edge-based streaming analytics framework that enforces differential privacy while processing high-throughput data streams with low latency.
Why notable — Privacy-preserving analytics at the edge is increasingly required by regulation and user expectation; this system shows it can be done at practical streaming throughputs.
LCL+: a Lock Chain Length-based Distributed Deadlock Detection and Resolution Service Built for OceanBase
Zhenkun Yang, Chen Qian, Xuwang Teng, Fanyu Kong et al.
TL;DR — LCL+ detects and resolves distributed deadlocks in the OceanBase database by tracking lock-chain lengths across nodes, achieving low overhead with fast detection latency.
Why notable — Deadlock detection in large-scale distributed databases is an unsolved production problem; this paper presents a battle-tested algorithm deployed in a major commercial system.