计算机学院系列讲座菁英论坛第26期——Next-Generation Load Balancing for Distributed Systems with Programmable Networks

报告题目(Title)：Next-Generation Load Balancing for Distributed Systems with Programmable Networks

时间 (Date & Time)：2024.04.02 14:00 – 15:00pm

地点 (Location)：北京大学第二教学楼211（燕园校区）

主讲人 (Speaker)：Dan R. K. Ports (Microsoft Research)

邀请人 (Host)；Xin Jin

报告视频会议 (Zoom Link)：ID 815 2852 529, Password 835276

报告摘要 (Abstract)：

Network-level load balancers are a mainstay of cloud platforms, providing the gateway to a variety of cloud applications. New programmable networking technology, including programmable switches and smart NICs, makes it possible to implement these load balancers with dramatically higher efficiency than existing software solutions, offering increased flexibility for custom, application-specific load balancing logic. In this talk, I will describe three new systems that leverage this flexibility to solve fundamental problems for distributed systems.

First, Pegasus addresses load imbalance in distributed storage by using an in-network coherence directory for load balancing, achieving over 10x throughput improvement for highly-skewed workloads. Capybara enables microsecond-scale TCP migration in data centers, facilitating fast failover and better load balancing with a 12x reduction in migration host latency. Lastly, Beaver introduces a practical partial snapshot protocol for cloud services, ensuring causal consistency under external traffic interference.

主讲人简介(Bio)：

Dan R. K. Ports is a Principal Researcher in the Systems Research Group at Microsoft Research, where he leads the Prometheus project — a long-term collaboration between MSR, academic research groups, hardware vendors, and product groups. Specializing in distributed systems and networking, Dan works across the entire systems stack, focusing on leveraging emerging datacenter technologies like programmable networks. Recent efforts within the Prometheus project have emphasized enhancing the performance, flexibility, and applications of cloud-scale load balancers. He received his Ph.D. from MIT in 2012 and previously served on the faculty at the University of Washington. Dan's contributions to the field have been recognized with best paper awards at NSDI and OSDI.

欢迎关注计算机学院微信公众号，了解更多讲座信息！

北京大学计算机学院