gMig: Efficient GPU Live Migration Optimized by Software Dirty Page for Full Virtualization
Efficient Shuffle Management with SCache for DAG Computing Frameworks
PPoPP 2018: 23rd ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming
Sat 24 - Wed 28 February 2018 Vösendorf / Wien, Austria
Our autumn outgoing’s destination is Xitang & Wuzhen in Zhejiang province. Everyone just enjoyed themselves and took a break from computer.
Kai Chen is an Assistant Professor with Department of Computer Science and Engineering, Hong Kong University of Science and Technology. He received his PhD in Computer Science from Northwestern University, Evanston IL in 2012. His research interests include networked systems design and implementation, data center networks, and networking support for cloud, big-data and deep-learning/AI systems. He values practical system research, has published in various top venues such as SIGCOMM and NSDI, and is now actively working with big IT companies such as Tencent, Huawei, Microsoft, Nvidia, etc., to translate his research into practice.
The link speed in production datacenters is growing fast, from 1Gbps to 40Gbps or even 100Gbps. However, the buffer size of commodity switches increases slowly, e.g., from 4MB at 1Gbps to 16MB at 100Gbps, thus significantly outpaced by the link speed. In such extremely shallow-buffered datacenter networks, prior TCP/ECN solutions suffer from either excessive packet losses or significant throughput degradation. In this talk, I will introduce BCC (Buffer-aware Congestion Control), a simple yet effective solution to this problem. BCC operates based on real-time shared buffer utilization: when the buffer is abundant, BCC delivers both high throughput and low packet loss rate; when it becomes scarce, BCC maintain low packet losses at the cost of sacrificing a small amount of throughput. We have validated BCC’s efficacy in a 100G testbed and demonstrate its superior performance using extensive simulations.
Welcome to join SING Lab(Chen’s Lab):