AccuWeather made a small change that has had a big impact on its computing needs.
When you check the local weather forecast on your smartphone, the device sends a GPS coordinate of 10 to 15 decimal points. If you're standing in Central Park, the request is narrowed down to a few yards of your exact location.
But as you might guess, the forecast you get on the west side of Central Park isn't any different from what you get on the east side. "I'm good at weather," says Steve Smith, CIO of AccuWeather, a 49-year-old company that specializes in providing weather forecasts. "I'm not that good."
Smith's problem is that the servers in his data center treat those two nearby forecasts as two separate requests, even though the data people receive is the same. So the data center generates a new response rather than use a cached copy of the request.
The problem is made more acute by the fact that requests for AccuWeather data have exploded in the last year, thanks to mobile devices. In January 2010, AccuWeather was getting about 100 million data requests a day. In January of this year, 750 million, many of them automated requests from devices to keep weather apps updated.
One way Smith has dealt with this challenge is to put five simple lines of code on the edge servers run by one caching service provider, AT&T. That code recognizes when someone's sending a GPS coordinate to request weather, and it truncates the number to two decimal points--about 1 kilometer. AT&T looks at the request and, if it's identical to one recently made, serves a cached version of the content.
The benefit from those few lines of code: 300 million to 500 million fewer requests a day handled by AccuWeather servers.
That means AccuWeather's data center doesn't need to process those requests, and users get the data faster, since it comes from AT&T's nearby content delivery network servers instead of AccuWeather's Pennsylvania data center. That workload reduction, plus a data center upgrade that involved moving to Dell blades with VMware's virtualization software, has helped AccuWeather handle the explosion in mobile data without expanding its data center.
This experience has Smith thinking what else can be pushed out of the data center. "I still feel I do too much serving" of information, he says.
If you think of cloud computing narrowly--as moving applications to Amazon's EC2 service, for example--Smith's innovation points to a new reality. There will be many cloud models. Amazon's business model is impractical for most of AccuWeather's computing. Amazon charges for the quantity of data processed, as well as for the number of requests. AccuWeather's typical data request is a tiny 4-KB XML file, but it gets hundreds of millions of them a day. Cloud computing is going to evolve to fill these niches.
Smith and his IT team aren't just innovating on infrastructure. Much of their new development is going into apps for mobile devices. AccuWeather was one of six apps available on the first iPad when it debuted.
To keep up with the pace of opportunity in mobile services, Smith has software developers working directly with marketing and product development teams. That takes a different mentality than IT pros are used to, having to start developing as ideas take shape and adjust as demand and devices change.
"IT folks are structured by nature," Smith says. "I'm asking them to get outside the box--you're not going to have it all written down--and fly by the seat of their pants a bit. Oh, and by the way, your timeline was yesterday. You don't have the luxury of months and months."