Researchers at Cornell University and Microsoft have cooked up a recipe for a wireless data center. That's right, an entire data center that shuttles information among thousands of machines using not cables but thin air.
Their design throws out traditional switches and networking cables, but it also borrows from a very unusual source: the 19th century mathematician Arthur Cayley.
It turns out that Cayley's 1889 paper, On the Theory of Groups, gave the researchers a mathematical model for figuring out how to best connect their servers into a wireless network that could keep on chugging even if some of the servers failed.
In a research paper, set to be presented at a technical conference in Austin, Texas, later this month, the researchers talk about ways that 60GHz wireless devices could be tweaked to pump out very high-bandwidth flows of data between the different servers.
The paper is important because the big web companies -- including Google and Facebook -- are already stretching the limits of how quickly they can pump large amounts of information across their data centers. These companies are exploring new ways of transmitting data using optical wire lines, but their problems could also be solved with a wireless data center like the one proposed by Cornell.
Cornell's proposal could also significantly reduce power and cost in the data center -- another concern for the giants of the web and many other companies.
The paper seeks to solve these problems with a new type of hardware. It doesn't deal with ordinary servers and ordinary boxy server racks. Cylindrical racks would house pie-shaped servers, and this would facilitate communication not only within the rack but with outside racks as well.
But the real trick is figuring out a way to keep the whole thing running when servers on the rack fail. And that's where Cayley's 120-year-old graph theory comes into play.
"Caley's responsible for showing that we have very strong connectivity," says Hakim Weatherspoon, a professor with Cornell University who co-authored the paper. "So our wireless center can tolerate a very high level of server failure."
They call their creation the Cayley data center. It hasn't been built yet, but if it does get funded, Weatherspoon believes that it will keep on working until 14 percent of the racks or 59 percent of the server nodes fail.
Networking companies have been working on 60GHz networking products for a few years now. These 60GHz transceivers operate at a much higher frequency than the Wi-Fi network you use at home. That means they're speedier, but without the same range. By using a cylindrical rack design and reworking networking protocols, the Cayley researchers think they can cut down on outside interference and keep data pumping at about 10 gigabits per second. That's remarkable, considering that 60GHz devices are supposed to operate in the 2- to 7-gigabits-per-second range.
Instead of engaging in back-and-forth communication chitchat you'd see in a typical wireless device, one Cayley server would connect with another, and then blast data, firehose-style to another, before signing off and waiting to receive information. Servers would talk to other machines within the rack using a transceiver on tip of the pie-shaped servers, and they'd reach out to other racks using a second transceiver on the back. So each server would be able to route data to the small number of other servers that it is set up to communicate with. That means every server is a kind of mini-switch -- called a Y-switch -- and none of the server racks need traditional networking switches for communications.
Because the Y-switch is based on cutting-edge technology, it will cost more than a typical networking card at first, but without the switching and cabling costs, Weatherspoon believes that his data centers will use less power and cost much less to connect.
They've already done some initial tests to see if the wireless interference would prevent the Cayley data center from working. Apparently, it won't. And if they can get tech companies like Intel or Microsoft interested in sponsoring further work, they could build protyotypes. "We think that it is attractive enough to maybe have some companies go a little bit further than what we've done," Weatherspoon says.
