News: ZeroSwap and "Per-Bank Memory Bandwidth Regulariton for Predictable and Performant Real-Time Systems" Receive Best Paper Awards!
The RTAS 2026 TPC shortlisted five candidates for the best paper awards. The best paper selection committee evaluated these papers carefully on the quality of the research as well as the potential scientific/industrial/societal impact. The committee also considered the reviews and discussions from the TPC members.
The best paper award goes to:
☆ ZeroSwap: Minimizing Swap Overhead for Real-Time Multi-DNN Inference via SSD-based GPU Memory Extension Woosung Kang (DGIST), Filippo Muzzini (University of Modena and Reggio Emilia), Gianluca Brilli (University of Modena and Reggio Emilia), Jong-Chan Kim (Kookmin University), Jinkyu Lee (Yonsei University), Hoon Sung Chwa (DGIST).
The committee noted that this paper addresses an increasingly important problem viz., scheduling of modern DNN workloads. In particular, such systems face problems since memory between CPUs and GPUs becomes a limiting factor. The paper presents novel methods for memory-swapping using SSDs to relieve bottlenecks in the scheduling of such systems. Apart from the novelty and technical depth of the paper, the selection committee noted that the paper proposes a complete end-to-end solution that subsumes integrated hardware and software mechanisms to ensure success. The paper is very well written and includes extensive evaluations. This paper has the potential for both, short-term as well as long-term impact.
The best student paper award goes to:
☆ Per-Bank Memory Bandwidth Regulation for Predictable and Performant Real-Time Systems Connor Sullivan, Amin Mamandipoor, Cole Strickler, Heechul Yun (all: University of Kansas).
This paper tackles the hard system problem of bank-level interference in modern DRAMs that are integrated with multicore SoCs and presents an innovative solution that achieves more than 5x improvement over competing methodologies. They carry out extensive experiments of DRAM performance and demonstrate a conceptual attack that takes advantage of the constant bandwidth across generations. Their solution, a per-bank memory bandwidth regulator, solves these problems in an effective manner. The committee noted that their implementation on a realistic SoC and evaluation using real-world workloads makes this paper really stand out.
Congratulations to the authors on the well-deserved best (student) paper awards at RTAS 2026.