We seek to address fundamental rules that govern future computer systems and networking.
The Accelerated Computer Systems Laboratory engages in the intersection of operating system and networking.
Operating System for NPUs
Neural Processing Unit (NPU) is a hardware accelerator explicitly optimized for machine learning workloads. Unlike general-purpose CPUs or GPUs, NPUs feature specialized hardware architectures, such as private scratchpad memories, hardware message-passing, and mesh-connected Network-on-Chip interconnects. While these architectural choices deliver high efficiency for machine learning applications, they break traditional operating system assumptions and demand dedicated system software support for resource scheduling, memory management, and cluster networking.
Systems for DPUs
Modern DPUs and SmartNICs have evolved beyond basic offloading to process complex data paths directly in hardware. By leveraging various hardware accelerators in DPUs, we develop new network operating system that provide both efficiency and fairness.
Networking and Operating System Support for Disaggregated Datacenters
On disaggregated datacenters where compute and memory are physically decoupled across high-speed networks, traditional operating systems introduce severe software overheads by treating network I/O as inherently slower than local memory. Our research addresses this paradigm shift by designing hardware-aware system software built specifically for near-memory-bandwidth interconnects. We address the challenges by building clean-slate OS architectures.
For newcomer BS/MS/PhD/PostDoc students in 2027 Spring
We are recruiting researchers to work on networking/operating system research including DPUs, NPUs, and memory disaggregation systems. Please email me if you are interested.
DPU, NPU, 메모리 디스어그래이션 등 네트워크/운영체제 연구에 관심있는 사람을 모집합니다. 학부 연구생, 석사, 박사 등 어떤 형태든 좋습니다. 관심이 있으면 로 메일주세요.
If you are interested in our group, contact (Prof. Wonsup Yoon).