• 목록

Professor Jinkyu Jeong's lab (Scalable Systems Software Lab) presented two papers at top-tier international conferences in the field of systems software (OSDI '24, USENIX ATC '24)

In July 2024, the Scalable Systems Software Lab at Yonsei University, led by Professor Jeong, presented a paper at the 18th USENIX Symposium on Operating Systems Design and Implementation (OSDI '24), the top international conference in the field of operating systems.

The paper titled “Identifying On-/Off-CPU Bottlenecks Together with Blocked Samples” introduces the Blocked Samples technique, a key innovation that greatly simplifies application performance profiling in today’s increasingly diverse and complex computer systems. The research presents two application performance profilers, bperf and BCOZ, which leverage this technique. These profilers are valuable tools for optimization as they pinpoint bottlenecks that lead to performance improvements when addressed. The research team demonstrated the utility of these profilers by profiling and optimizing the performance of large-scale language model inference and NoSQL big data storage systems. Additionally, the Blocked Samples technique reduces application performance interference by up to 17 times compared to similar existing profiler tools.

Additionally, Professor Jeong’s research team presented a paper at the 2024 USENIX Annual Technical Conference (USENIX ATC '24), a flagship international conference in the field of system software, which was held alongside OSDI '24.

The paper titled “A Secure, Fast, and Resource-Efficient Serverless Platform with Function REWIND” identifies security issues caused by container (or sandbox) reuse techniques used to enhance performance in commercial serverless cloud platforms like Amazon Lambda and Google Cloud Functions. The paper introduces the REWIND technique, which addresses these security concerns while simultaneously improving performance and reducing memory usage. This approach selectively rewinds only the memory and file regions that could cause security issues after executing a serverless function within a serverless container. By doing so, it eliminates any residual user privacy data, ensuring security, while significantly reducing the memory usage required to maintain this security. The research team demonstrated that, across various real-world cloud workloads, the proposed technique maintains near-zero performance loss compared to less secure execution methods and reduces memory usage by more than half. 

Links:

Identifying On-/Off-CPU Bottlenecks Together with Blocked Samples, https://www.usenix.org/conference/osdi24/presentation/ahn

A Secure, Fast, and Resource-Efficient Serverless Platform with Function REWIND, https://www.usenix.org/conference/atc24/presentation/song