Networking at the Speed of Light
A good Internet experience requires responsiveness, especially with human applications like web browsing and interacting in real time with voice and video. Even tens of milliseconds of delay can affect perceptions of the “latency” lag and lead to significant reductions in website revenue. During his Center appointment, Professor Godfrey will work to establish speed-of-light latency as a grand challenge for the computer networking research community. The goal is to achieve responsiveness close to the underlying physical limits.
Professor Godfrey’s agenda includes conveying this long-term vision to the research community, conducting measurement studies to map out latency problems and opportunities, and developing new technologies to reduce latency.
First, he will develop an understanding of why latency is generally inflated by more than an order of magnitude beyond the ideal. He will build a comprehensive picture of each layer of the network architecture, including the physical location of fiber lines, routing paths within those lines, and transport protocols used to establish end-to-end communication. He will then compare the baseline ideal speed-of-light latency with measured latency and correlate the results with the various Internet Service Providers (ISPs) involved. A continual release of this data analysis is expected to promote competition among ISPs to reduce latency.
Two additional elements of this project will result in direct contributions. Professor Godfrey is exploring the use of redundant requests to achieve consistently low latency in Domain Name System resolution, storage services, and multipath routing, which will result in open-source software that accelerates web applications. He will also develop and release open-source software that makes control decisions based on real-time analytics of performance outcomes. While this software represents a radically new architecture, it works within existing protocols and requires changes on the sender side only.