Lunch will be served at 11:45 AM.
The amount of daCPUs have long served as the workhorses for processing network traffic, but they are increasingly struggling to keep pace with the rising communication demands in next-generation data centers. In response, specialized network devices—e.g., smart network interface cards (SmartNICs)—are emerging to offload network tasks from the CPU. However, hardware specialization introduces its own challenges: these SmartNICs can be difficult to program and deploy, and they raise complex system-level issues for extracting optimal performance.
In this talk, I will demonstrate a holistic re-architecture of the next-generation data center network stack—spanning hardware, software, and compiler toolchains—with the goal of 1) building abstractions to hide the complexities of specialized network devices, yet making them simple and easy to use, and 2) use hardware-software co-design to maximize the end-to-end performance that applications experience. I will begin with Alkali, a SmartNIC compiler framework that abstracts away hardware heterogeneity and complexity, enabling the "write once, run anywhere" paradigm. Next, I will introduce PANIC, a widely used open-source SmartNIC architecture that provides OS-like multi-application performance isolation on the NIC via novel hardware components. I will also briefly discuss Ringleader, a system that co-designs NIC hardware with the operating system to provide ultra-low request serving latencies.
Domain-specific hardware development is intensifying, with SmartNICs being just one example. I will conclude my talk by discussing the broad opportunities for crafting novel abstractions and harnessing automated co-design to shape the next-generation data center.
Jiaxin Lin is a Ph.D. candidate at the University of Texas at Austin. Her research spans computer networks, systems, and architecture. She is widely recognized for her influential work on programmable NICs for which she won the Google and Meta PhD Fellowships in Computer Networking. In 2024, she was chosen as an MIT EECS Rising Star. Her overarching goal is to design innovative software and hardware systems that fully leverage the semantics and capabilities of advanced network devices and interconnects.