Google: Instant Search Is Expensive
The new Google Instant feature is popular with users, but imposes a heavy load on the company’s technical infrastructure
Google Instant, the predictive search technology the company launched last month to serve users results without making them hit the enter button, is popular but costly.
Jonathan Rosenberg, Google’s senior vice president for product management, asserted on the company’s third-quarter earnings call on 14 October that not only was Instant not created to help Google make more money, but that “from a resource standpoint, it’s actually pretty expensive”.
Additional resources
Rosenberg also declined to explain what resources triggered those costs, so eWEEK asked Google.
A company spokesperson confirmed that Google had purchased additional computer servers to deliver the results, but declined to say how many new machines and what the cost was to not only build Instant but keep it pumping out queries with each tap of a keystroke as it does today.
“The cost of search has steadily increased over the years as we develop new innovations to serve users,” the spokesperson said. “We can’t give you a specific number of machines we’ve added to support this launch, other than to say this is technically demanding for our infrastructure.”
This is par for the course for Google, which has historically declined tto discuss how many servers they employ to fuel its data centres all over the world.
Hardware is hardly the only resources Google requires for Instant, which serves an average of 5 to 7 times more result pages for some queries. The company uses some nifty software and packet traffic tricks. Those are a bit more public.
Technical tricks
For example, the Google spokesperson said Google is reducing demand on servers by keeping the frame of the results page the same while dynamically generating new results within that frame.
This is what keeps Google Instant appearing so smooth and seamless as results appear and disappear from the screen with each keystroke. Google also built systems to control the rate at which Instant shows results pages in proportion to how relevant the pages are likely to be.
Also, the company’s search infrastructure team deployed new server caches that can handle high request rates while keeping results fresh as it the search algorithm crawls and re-index the web.
New client caches in Google’s infrastructure include user-state data to track results pages already shown to a given user so as not to this re-fetch the same results.
Meanwhile, the Instant team is working hard to port the Instant technology from the desktop to mobile devices. Rosenberg said on the call that users can expect Instant to come to Google Android, RIM Blackberry and Apple’s iPhone later this autumn.
“It’s relatively soon. Sometime this fall. Fall lasts a little longer in California though,” he joked.
