CNS Core: Medium: The Synchronous Data Center

University Of Pennsylvania, Philadelphia PA

Investigators

Abstract

Most major web services - such as Amazon, Google, or Facebook - employ massive data centers that can each contain hundreds of thousands of computers. Today, these data centers are almost always built using an asynchronous design, with few or no assumptions about how long it might take for two computers to communicate, or when exactly a computer might be able to complete a given task. The 'asynchronous design' has been the standard since the very beginning of distributed computing; it is the one that is taught in textbooks, and it is rarely questioned. However, the design is not as simple as it may at first appear: in practice it often leads to low efficiency, unpredictable timing, and many other serious problems. This project proposes a radical new solution--synchronous networks-- that intrinsically eliminate these problems. At a high level, this approach involves creating a detailed choreography that describes what each computer should do and when. While synchronous designs can be more complex, they also come with considerable potential benefits, including higher efficiency, better predictability, and support for new kinds of services. In addition, they can potentially sidestep some of the most difficult problems with asynchronous designs. This project will develop the management architecture and necessary algorithms and other components for this new model and explore its performance in a network testbed and through simulation. The project is leveraging insights from cyber-physical systems, which have been using synchronous designs for decades, albeit at much smaller scales and in a very different environment. It is developing 1) a prototype of a new, fully synchronous computing platform; 2) new, scalable coordination mechanisms and scheduling techniques; and 3) a set of synchronous building blocks for applications. The project is using web search as a case study to demonstrate the end-to-end benefits of the synchronous approach. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.