WolframAlpha Works Through Tech Gliches
The WolframAlpha search engine made its debut Friday and continued testing through the weekend. The team brought up the engine live through an Internet feed, though it seemed like everything that could go wrong did.
A few glitches in the live broadcast set the team scrabbling to make excuses for a late start. A tornado watch in Champaign, Ill., the official launch site, threatened to interrupt takeoff. Officially, the site goes live Monday. On occasion an error message would appear stating "I'm sorry Dave, I'm afraid I can't do that..." to notify the site had exceeded its test load because some of the servers processing queries were not working correctly.
Although the WolframAlpha team sent a warning on the company's blog that the site may go down periodically through the weekend, creator Stephen Wolfram, a British computer scientist, made the "computational knowledge engine" available for anyone to plug in questions.
Fifteen hours into the launch, Wolfram posted a video on the company's blog saying that the engineers fixed the stat servers. They now worked properly. This means engineers bringing up the search engine began to see the queries being processed on the servers.
At about 1 p.m. central time Saturday, the servers began processing about 80 queries per second. Roughly 70% of the people asking questions got answers, though some were incomplete, according to the video log. Other problems occurred, too, such as real-time data feeds, such as weather, were not getting propagated out to servers properly.
WolframAlpha provides a new way to discover information. Unlike Google, Microsoft and Yahoo, the search engine pull stats from trillions of pieces of information sourced from thousands of Web sites, libraries and academic journals, and checked by experts. A source list appears at the bottom of the search queries. The search engine focuses on music, science, health and medicine, transportation, engineering socioeconomic, words and linguistics, and people and history. Results appear in text, words and graphs.
In a brief 10-minute test of one cluster made live to the world last week prior to launch, about 3,000 people found there way in to search on the site. Roughly 18,000 queries ran by in that short time frame.
Those queries were used to test system loads. One of the largest clusters, about 3,800 CPUs sits at the headquarters occupying most of the power consumption in a very large data center. It's one of five centers that house about 10,000 CPUs able to handle thousands of queries per second.