Keynote Talks

Haibo Chen

Distinguished Professor of Shanghai Jiao Tong University; IEEE Fellow

Talk Title:

Microsecond-scale Datacenter Computing with RDMA: Characterization, Optimization and Outlooking

Abstract: Many online services are both latency-critical and throughput-oriented, which require datacenters to provision resources elastically and at microsecond-scale. Meanwhile, datacenters are now increasingly equipped with low-latency hardware like RDMA-capable SmartNICs as well as heterogenous on-chip and off-chip computing devices like CPU, GPU and NPU. In this talk, I will first present a characterization of emerging RDMA-capable NICs and SmartNICs and their interactions with computing devices, which uncovers many important performance features not systematically understood before. Based on this, I will summarize a set of optimization guidelines to efficiently harmonize RDMA with computing devices to provide microsecond-scale (tail) latency yet still high throughput. Then, I will present a set of case studies utilizing such guidelines to achieve significantly much better performance over state-of-the-art. Based on our experiences, I will outlook a set of open challenges and opportunities that may open new spaces for further evolution.

Speaker Bio: Haibo Chen is a Distinguished Professor of Shanghai Jiao Tong University, where he directs the Institute for Parallel and Distributed Systems (IPADS). His main research areas are operating systems and distributed systems. He received Best Paper Awards from ASPLOS, EuroSys and VEE, Test of Time Award from DSN, Best Paper Honorable Mention and Research Highlight Award from SIGMOD, Honorable Mention of The Dennis M. Ritchie Thesis Award (Advisor) from SIGOPS. He currently serves on the editorial board member and co-chair of Special Sections of Communications of the ACM, Program Committee of SOSP 2023/OSDI 2024, PC co-chair of EuroSys 2025, and the inaugural technical steering committee chair of OpenHarmony, an open-source operating system deployed on hundreds of millions of devices. He is an IEEE Fellow and an ACM Distinguished Member.

John Kim

Professor in the School of Electrical Engineering at KAIST

Talk Title:

Domain-Specific Interconnection Networks in the Era of Domain-Specific AI Supercomputer

Abstract: Domain-specific architectures are hardware computing engine that is specialized for a particular application domain. As domain-specific architectures and AI supercomputers are widely used, the interconnection network can become the bottleneck for the system. In this talk, I will present the role of domain-specific interconnection networks to enable scalable domain-specific architectures. In particular, I will present the impact of the physical/logical topology of the interconnection network on communication such as AllReduce in domain-specific systems. I will also discuss the opportunity of domain-specific interconnection networks and how they can be leveraged to optimize overall system performance and efficiency. As a case study, I will present the unique design of the recent Groq software-managed scale-out system and how it adopts architectures from high-performance computing to enable a domain-specific interconnection network.

Speaker Bio: John Kim is currently a full professor in the School of Electrical Engineering at KAIST (Korea Advanced Institute of Science and Technology) in Daejeon, Korea. John Kim received his Ph.D. from Stanford University and B.S/M.Eng from Cornell University. His research interests include computer architecture, interconnection networks, security, and mobile systems. He has received a Google Faculty Research Award, Microsoft-Asia New Faculty Fellowship, and is listed in the Hall of Fame for ISCA, MICRO, and HPCA. He has also worked on the design of several microprocessors at Intel and at Motorola.