Facebook Research awards novel research from USI
Institutional Communication Service
When referring to Facebook, one is most likely to think of it as the popular social network used every day by billions of people across the planet. Not everyone may know, however, that the leading global firm – like others in the Big Tech league, e.g. Google or Amazon – invests significantly in frontier academic research by inviting academics to submit research proposals in “specific areas that align with its mission of building community and bringing the world closer together”. In October 2019, Facebook launched a call for proposals in the area of Distributed Systems and, on February 28, 2020, announced the eight winners of the grants, which include the project submitted by Patrick Eugster, Full professor of Informatics at USI, together with postdoctoral researcher Pierre-Louis Roman, also at the USI Faculty of Informatics.
The USI proposal is the only one hailing from outside the United States, sitting alongside projects from renowned universities such as the MIT, Princeton, and Carnegie Mellon. The eight winners were selected among 63 proposals from 12 countries and 50 universities, thus underscoring the importance of academic engagement in the industry. The winners will attend the Core Systems Faculty Summit and Data Center Tour at Facebook in May 2020, where they will have the opportunity to discuss their proposals with the research community.
Towards hybrid asynchronous/synchronous distributed systems is the name of the project proposal submitted by Prof. Eugster and Dr. Roman. As Prof. Eugster explains, “A major pain point in applications like Facebook, which executes its services across many distributed host servers, is the need to coordinate among these hosts to avoid, for example, that a given task is performed multiple times or, inversely, never performed. The reason why coordination is hard is that datacenters in which hosts run are assumed to be asynchronous distributed systems, which means that in theory there is no upper bound on the time needed to communicate between two hosts, and there is no upper bound on the time it takes one host to compute a response for another host. At the same time, hosts can fail, yet without synchrony it is impossible to tell a slow host from a crashed one. As a result, current protocols for coordination are complex and costly. With our project we propose a novel architecture for speeding up distributed applications that, one the one hand, concentrates synchrony guarantees to only certain “parts'' of such distributed applications, e.g. coordination tasks and, on the other, leverages recent developments in commodity hardware and software to achieve time bounds on communication and computation bounds whose tightness is unprecedented”.