Wolfram Alpha Braces for Overload
The computational knowledge engine expects a flood of users.
Wolfram Alpha, a “computational knowledge engine” that answers queries by tapping databases and running calculations instead of searching the Web, went live on Friday night and managed to avoid crashing over the weekend and on Monday, the first full business day of operation.
The engine was launched on schedule despite failing a large-scale traffic simulation test earlier on Friday. The site quickly began receiving 100 hits per second, a figure that climbed steadily over the weekend. As of early Monday morning, the company said that Wolfram Alpha had processed a total of 13.7 million queries from six continents and had received 27,000 responses to requests for feedback.
Some people used the engine to view nearly real-time information about a small earthquake that struck Southern California on Sunday, the company said. (Google can now provide the same information using news feeds from the U.S. Geological Service.) Meanwhile, Facebook users suggested search terms for each other to try. The site did not always function as planned, however. Queries were sometimes answered with error messages citing overloads. And today is expected to bring even more traffic.
Whereas search engines like Google use algorithms to find Web pages that are deemed most relevant to a particular query, Wolfram Alpha interprets the query, accesses its own databases, and calculates answers. If a user enters a term such as “GDP of France and Mexico,” it returns not a list of Web pages but relevant data compared and organized into tables, charts, and graphics. Data is stored in databases maintained by Wolfram Research and computed and visualized using the company’s math software package Mathematica. (See a comparison of how Google and Wolfram Alpha treat the same queries here.)
Stephen Wolfram, the 49-year-old founder and CEO of Wolfram Research, gave Friday’s events a theatrical touch by deciding to webcast the launch, which took place inside the company’s control center in an industrial building in central Illinois. At 10:35 P.M. local time, with the cameras taping, he mouse-clicked a tab that said “Activate,” and the company’s data centers began processing requests.
In recent weeks Google has announced two new services with some similarities to Wolfram Alpha. The first allows visualization of public data, starting with census and labor statistics. The second, Google Squared (due out from Google Labs later this month), processes information from Web pages to create tables of data on a broad range of subjects.
In an interview Friday, Peter Norvig, Google’s research director, said that Google is working on various ways to find and present useful numerical data. He said that he hadn’t seen the Wolfram site in action and would “have to see the reaction” from searchers once it goes live. “Maybe having him out there will push us to release more, faster–I don’t know,” he said.
The launch of Wolfram Alpha was certainly not as disastrous as Cuil’s debut last year, but it was not without glitches. In the initial hours, displays in the data center were not accurately logging how many queries the site was receiving or where they were coming from. For clues, Wolfram Research engineers looked at chat postings accompanying the webcast. (People reported “working in Bloomington,” “not working in Madrid,” and so on.) “We’ve got crowdsourced logging!” Wolfram joked.
The service will require more computing power than is used for typical Web searches, says Samer Diab, chief operating officer of Wolfram Solutions, a unit of Wolfram Research. The company partnered with Dell and R Systems to run the site with two supercomputers and three data centers.
“Wolfram Alpha, by its nature, is going to be a very computationally intensive website,” Diab says. “There is a huge amount of work to put together behind the scenes that will support the volume of calculation that will end up happening.”