The company has been manufacturing its own networking equipment for the past seven years

Apr 18, 2012 07:10 GMT  ·  By

Google has been manufacturing custom networking equipment for use in its own networks since 2005. The company never officially confirmed this, but there was rumor, back in 2005, that they were - and subsequent specific job listings with Google have provided enough confirmation for industry insiders.

Two years ago, a study done by Arbor Networks concluded that if Google were an internet service provider (ISP), it would be the second largest ISP in the world.

To get an idea of what that would be like, we just need to take a look at the largest ISP on earth. It’s called Level 3 and it provides internet and networking services for 2700 major corporations, in 450 markets around the world.

Of course Google is not an ISP, but they do a lot of data moving across their networks around the world.

In the networking business, a network designer builds for the network's “peak.”

That means you don’t build your network for the average amount of traffic, but you build it to ensure that when the amount of traffic reaches peak of the day, week, or month etc., your network will still be online, able to service the requests that are issued and will most definitely not crash.

This translates into a lot of over-provisioning and generally means that your network’s capacity is much greater than what’s actually used most of the time.

The industry experts consider that an average 30 to 40 percent network utilization is quite an achievement, but Google’s Urs Holzle says the company’s networks are getting closer to 100% usage without any traffic disruptions.

Back in 1999, Urs Holzle was a professor at UC Santa Barbara and, being surprised by what a small startup in Palo Alto was doing, he took a year-long leave and started working for Google, designing the company’s networking infrastructure.

It goes without saying that Mr. Holzle has never left the company, and now he’s heading Google’s implementation of Software Defined Networking (SDN).

SDN networks are characterized by the fact that the system always knows where each packet of traffic data is, despite the fact that data packets are just as independent as they are in current networking implementations.

The obvious advantage is that the system can command some non-urgent packets to use possibly slower detour routes to make way for essential / urgent packets that need to be delivered right away.

Taking into consideration the fact that the detour routes might actually be free and available to swallow the traffic, you will get a double bonus: all the essential / urgent packets are delivered in the fastest way and even non-urgent traffic arrives ahead of schedule, making a satisfying impression on the final user.

Therefore we now have a network that can handle “the peak” at any time and that won’t actually let any resource go unused; instead it will put to work less utilized resources to improve the end user services and command the central links in such a manner that they’re always available to deliver huge amounts of data with no traffic jams.

This network model is being developed by two teams of scientists at Stanford and University of California at Berkley and it’s called OpenFlow.

Google became one of several organization to be part of the Open Networking Foundation, with other members including the likes of Yahoo, Microsoft, Facebook, Verizon and Deutsche Telekom, and others that are working to make this model popular.

None of the other members are using such a wide implementation of the technology in their internal networks like Google, but they aren’t manufacturing any custom networking equipment either.

Google says that it will eventually compete with the Cisco and Juniper, but also mentions the fact that if the two networking giants start adopting the SDN OpenFlow model, Google will most likely be a client and not a competitor.