It took decades before a bee algorithm assisted engineers in making Web hosting services more efficient.
By Rick Paulas
(Photo: daveograve/Flickr)
My partner collects stuff. The kind word for this is “conservationist,” the mean one is “pack-rat.” And while I occasionally get flustered living among what some may consider clutter, I have to consistently hand it to her: The stuff she saves gets used. Maybe not for a week, maybe not for a year, but when the day comes that we want to decorate for Halloween, it’ll be a good thing we kept those fake cobwebs around.
Scientific discoveries work the same way. You see a problem, and so you work it out, not quite knowing how it may fit in with the rest of the world. Sometimes it takes years until the science finds a real-world application. For the Honey Bee Algorithm, which has made Internet Web hosting services dramatically more efficient, it took decades.
Back in 1988, Georgia Tech engineering professor John Hagood Vande Vate was on his way to work when he heard an NPR interview with Cornell University honeybee researcher Tom Seeley. As the recap at the Golden Goose Award website tells it: “Seeley explained how the thousands of nectar foragers in a honeybee colony manage to distribute themselves wisely among local flower patches without any central authority.” This struck a chord with Vande Vate, who wondered if bees could make consultants of sorts in determining the quickest and most efficient routes in the transport of data.
Vande Vate and his fellow engineers at Georgia Tech — including John J. Bartiholdi III and Craig Tovey — spent two years developing a theory before proposing a collaboration with Seeley. The newly expanded group then spent another year refining a set of equations to predict how the bees would distribute themselves to best accumulate nectar before returning to the hive. Then, it was time to head out into the field to test their model.
“It was BYOB,” Tovey says. “Bring your own bees.”
They brought a colony of 4,000 honeybees to Cranberry Lake Biological Station in upstate New York, taking great pains to provide each with its own unique ID code of two digits and two colors. They also set up a number of “nectar sources” that would allow them to examine how each forager bee would know where to go, how long to spend there, etc. The information was delivered through a dance move the bees performed when they returned to the hive, a figure-eight “waggle dance.”
“It was BYOB,” Tovey says. “Bring your own bees.”
“The waggle dance tells other bees what patch it came from and what flowers are there,” Tovey says. “Other bees follow the dance and get recruited to go.”
This simple way of self-allocation solved any problem presented to the hive’s members, whether it was the closing of a flower patch, the discovery of a new one, or even bees dying in the middle of their work. Tovey knew that flexibility could have real ramifications in the non-bee world. “Imagine all the employees in one of your departments got sick and didn’t show up to work,” Tovey says. “That workplace wouldn’t function. But for the bees, it’s not a big deal. They just automatically reallocate.”
The group used the field work to tweak their original model, but then they had an even bigger problem: What could they use it for? “It had all these beautiful features,” Tovey says. “I thought it ought to be good for something.”
The Honey Bee Algorithm sat unused for a decade until, as Tovey tells it, a tall man walked into the office looking for help. He was Sunil Nakrani, a graduate student from the University of Oxford that had previously worked for IBM. His problem involved how the company allocated its servers. It rented space to customers, each of whom paid a percentage depending on how often the servers were being used. This structure meant that unused servers were not making money, but it also meant that their inaction could potentially be tapped to speed up traffic. Nakrani wanted to know if there was a better way.
There were a few problems in the algorithmic case Nakrani presented. First, traffic levels are unpredictable. “If a hurricane is threatening, Weather Channel traffic can go up by a factor of 1,000 in a couple hours,” Tovey says. And each website had different usage-rates: a banking website will be used for a different length of time than someone visiting a blog. “I thought, oh, that’s like honeybees,” he says. “Because they have to figure out how many bees to send to each flower patch.” But the clincher was when Nakrani told Tovey that the algorithm would need down-time built in to allow the server to change from one website to the next. It’s not as simple as flipping a switch; an online bank has way more security requirements than a blog. “That [downtime] is true with honeybees too,” Tovey says.
The waggle dance, you see, isn’t all that precise. It typically takes three tries for a honeybee to correctly register the message sent by the waggler. The downtime needed to switch servers from one action to the next, then, was built into nature. “I’ll never forget listening to him and in my mind seeing the shape of the curves,” Tovey says. “I told him, this problem is so much like the honeybees.”
Using the algorithm instantly sped up the time needed for customer transactions, allowing the company to earn 20 to 25 percent more income. Since that connection was made, the academic paper describing the algorithm has been cited more than 230 times by researchers. And because the engineering team put the entire discovery into the public domain, Web hosting services have implemented this algorithm into their own products, creating more efficient routes for their traffic, and essentially making a faster Internet for everyone. (Not to mention earning the hosts tons more money.)
Tovey has a few other algorithms in his back pocket that he’s waiting to find some use for. He’s been studying how fire ants build rafts and towers out of their own bodies in the hopes it may lead to helping search-and-rescue robots realign themselves in different environments. And his research on ant colonies — where millions of workers perform tasks with no central management — is still waiting to find a home.
In the meantime, Tovey and company and other scientists around the world will continue doing what they’ve always done: Solving problems that don’t exist just yet. This is worth thinking about the next time politicians try to defund obscure research just because they don’t see the point.