During his keynote speech at the Open Networking Summit in Santa Clara on Tuesday, Google executive Urs Holzle detailed the company's work with software-defined networking (SDN) from its earliest stages.
Google's main challenges with networking derive from its need to maintain two major backbones: one to handle external Internet user traffic and another for that between Google's internal data centers. The latter, which also happens to be the larger network, has been fully supporting OpenFlow for about the past six months, and is the result of an ongoing project launched in the spring of 2010, says Holzle, a Google fellow and senior vice president for technical infrastructure.
Holzle says Google's top priorities when it first started its SDN project were intensive pre-deployment and testing. By mid-2011, Holzle says Google had activated a simple SDN, and by early 2012 the company was running a full production SDN in one of its data centers. In January, Google was able to deploy centralized traffic engineering of its new SDN, which Holzle cites as a main objective of the project because it gives the company the ability to target network failures more quickly and gain concentrated control over the network.
Google can also now simulate its entire backbone network in software for isolated testing, Holzle says. Other benefits of the project include the ability to separate hardware from software so upgrades and new products can be chosen based on specific criteria, as well as cost benefits that Holzle says cannot be quantified yet.
While Holzle says Google has been "happy with the stability" of this new network, he acknowledges that there have been outages. None were out of the ordinary, though, and Holzle says he considers the configuration issues that caused those crashes "just a maturity thing."
Similarly, a few lingering challenges continue to drive Google's push for further standardization, including the amount of time it takes to program for a large network and determining how to configure the hardware that remains needed.
While Google continues to tinker with SDN technology on its data center network, it may not be long until the everyday Internet user reaps SDN benefits, too, Holzle says. Google will naturally have to wait for network equipment vendors to develop SDN-compatible offerings, but Holzle says he "would be disappointed if two years from now we didn't have some OpenFlow control in [the external network]."
"The time is right and it's not that complicated. We didn't have a huge team on it, so I'm sure that vendors can do that too in a finite amount of time," Holzle says.
At that point, SDN could have a significant impact on network performance at the end user level, Holzle says. Citing the example of group video chat on Google+, he says an SDN would afford the ability to establish a temporary virtual private network solely for the purpose of supporting that conversation, eliminating such performance issues as bandwidth limitations.
Altogether, Holzle believes that the work Google is doing on its data center network could translate to even greater benefits when it reaches the more complex public-facing network, as bringing simplicity to a more complicated environment is where SDN is most valuable.