Google Reuses Existing Infrastructure To Speed Up Searches
October 25, 2013

Google’s Cache Centers Deliver Data Quickly

Michael Harper for - Your Universe Online

This summer Wired published a report claiming Google uses Global Cache Centers around the world to quickly deliver up to 25 percent of the world’s Internet traffic. Now researchers from the University of Southern California (USC) say these Cache Centers have increased the number of locations from which Google serves up data by 600 percent. These servers operate under a partnership with more than 850 ISPs all over the world, though according to this summer’s Wired article, the ISPs aren’t keen on discussing the details of these partnerships.

The Cache Centers are used to serve up the most requested data to users through the ISP rather than direct all traffic to Google’s data centers. This means popular YouTube videos can be stored in cache memory and delivered quickly. According to USC PhD student Matt Calder, lead author of the study, Google has expanded these networks and has begun using them to deliver more than just popular videos. Calder and team will deliver the study at the SIGCOMM Internet Measurement Conference in Spain on October 24.

“Google already delivered YouTube videos from within these client networks,” explained Calder. “But they’ve abruptly expanded the way they use the networks, turning their content-hosting infrastructure into a search infrastructure as well.”

According to their research, passing requests through an ISP installed Cache Center may sound counterintuitive to delivering fast search results, but this actually increases the speed at which results are delivered. By maintaining a steady connection between the Cache Centers and Google’s main servers, the connection between the two remains fast and stable. Additionally some requests can be split into packets and handled on both the ISP and the server side. Searches which in part include information stored on the Cache Center, for instance, might get split into multiple packets and handled by both pieces of the infrastructure at once.

Calder his and team set out to map where these packets are sent, how they’re handled, and how quickly they’re delivered. They developed a method to monitor these packets and say they identified where the data centers live as well as the types of relationships these centers have with both servers and clients. Coincidentally, the team began monitoring these data centers and relationships as Google began to switch to their ISP server-side model of delivering data. This allowed them to track the changes as they were happening and compare the new results to the old.

“Delayed web responses lead to decreased user engagement, fewer searches, and lost revenue,” said Ethan Katz-Bassett, assistant professor at USC Viterbi. “Google’s rapid expansion tackles major causes of slow transfers head-on.”

This summer Wired said Google was on a “building binge,” and had planted data centers on at least four continents. While this build out received a fair amount of attention, their quiet plans and deals with ISPs got much less attention, and while they’re not the only company to employ this kind of technology, they might be the largest to do so.