Beehive - A Scalable Disaggregated Memory Runtime Exploiting Asynchrony of Multithreaded Programs

Watch this 13-minute conference talk from NSDI '25 that presents Beehive, a novel memory-disaggregated framework designed to improve remote access throughput by exploiting asynchrony within multithreaded programs. Learn how researchers from the Institute of Computing Technology at the Chinese Academy of Sciences, Peking University, Huawei Cloud, and Shanghai Jiao Tong University address the tension between programming productivity and performance in microsecond-scale I/O fabrics for disaggregated memory systems. Discover how Beehive maintains the intuitive multithreaded synchronous programming model while automatically transforming code into pararoutine-based asynchronous code through the Rust compiler, eliminating the poor data locality and scheduling overhead associated with traditional thread switching approaches. Explore the technical innovations that enable Beehive to outperform state-of-the-art memory-disaggregated frameworks including Fastswap (4.26× improvement), Hermit (3.05× improvement), and AIFM (1.58× improvement), making it a significant advancement in disaggregated memory runtime systems for high-performance computing applications.