An image that shows the optical switching in the subnanosecond range. Photo credit: Kevin Price / Detail Design Consultants / UCL
A new technique that synchronizes the clocks of computers in less than a billionth of a second may remove one of the hurdles to providing all-optical networks and potentially lead to more efficient data centers, according to a new study from University College London and London, according to Microsoft.
Data centers, comprising tens of thousands or hundreds of thousands of connected servers, are the underlying technology that enables everything we do online, from storing movies and photos to serving web pages and online services. However, you are facing a rapidly increasing demand, with server-to-server traffic increasing 70% each year, which is increasingly difficult to manage with existing technologies. All-optical networks that use light to both send and route data are a promising alternative. However, their viability was limited as each server had to continually adjust its time to match the incoming data, resulting in lower overall performance.
The study, published in Natural electronicsshows that by synchronizing the clocks of all connected servers over fiber and programming hardware to store clock phase values so that the clock time does not have to be checked again, the time to “restore” the clock could be virtually eliminated.
PhD student Kari Clark (Optical Networks Group, UCL Electronic & Electrical Engineering, winner of the EPSRC Connected Nation Pioneers competition), lead author of the study, said, “Our research is making optical switching possible for the data center for the first time by providing a solution for the problem of clock synchronization. It has the potential to transform the way computers communicate in the cloud and to use key future technologies like the Internet of Things and artificial intelligence cheaper, faster and less electricity. ”
So far, cloud providers have been able to meet the rapid growth in demand by relying on Moore’s law for networking. Integrated circuits with electronic switches double their data transmission speed about every two years for the same cost and performance. However, the sustainability of this trend is increasingly being questioned as it is difficult to keep making silicon transistors smaller and faster.
Dr. Hitesh Ballani and Dr. Paolo Costa, researcher at Microsoft Research Cambridge and co-author of the study, added: “With the expected slowdown in Moore’s Law and the ever-increasing cloud traffic, all-optical networks are an attractive technology that has remained elusive. We are very pleased about this collaboration with the UCL Optical Networks Group, which began in 2016 with Kari’s internship in our laboratory and developed into a multi-year journey as part of the Optics for the Cloud Research Alliance. Although there is still a long way to go, this technology brings us one step closer to the vision of a purely optical data center. ”
Dr. Zhixin Liu (Optical Networking Group, UCL Electronic & Electrical Engineering), lead author of the study, said, “We began this work by investigating how future cloud services can be supported beyond the end of Moore’s law. By bringing together the top minds of cloud operators and research in the field of optical communication, we propose a future-proof alternative with optics that will help data centers cope with long-term demand. ”
The team worked with researchers from Microsoft Research Cambridge to develop a prototype and found that its technique, known as clock phase caching, clocks thousands of computers in less than a billionth of a second, or the time it takes for light is needed to synchronize can travel 30 cm in the air.
The authors showed that reducing clock recovery time to below a nanosecond resulted in a significant increase in optical switching performance compared to modern solutions, making it practical for data centers and unlocking its full potential.
Reference: “Synchronous subnanosecond clock and data recovery for optically switched data centers with clock phase caching” by Kari A. Clark, Daniel Cletheroe, Thomas Gerard, Istvan Haller, Krzysztof Jozwik, Kai Shi, Benn Thomsen, Hugh Williams, Georgios Zervas and Hitesh Ballani, Polina Bayvel, Paolo Costa and Zhixin Liu, June 22, 2020, Natural electronics.
DOI: 10.1038 / s41928-020-0423-y
This work is funded by the Optics for the Cloud Research Alliance, the EPSRC program Grant TRANSNET (EP / R035342 / 1), and the EPSRC New Investigator Award (EP / R041792 / 1).
