Traditional networking gear just can’t keep up with the demands of today’s modern data center. At least that was the message delivered from executives of both Google and Microsoft at this week’s Open Network Summit.

The problem is that the traditional network switches and controllers built by companies like Cisco Systems Inc. just don’t have the ability to scale to meet the needs of cloud vendors who operate multiple sprawling data centers across the world. In both cases, Google and Microsoft have had to build their own software-defined networks instead, and in the former’s case, it’s been doing so for the last ten years.

Google provided some details about its project on Wednesday at the Open Network Summit and in a blog post. Its current network, which powers all of the company’s data centers, has a capacity of 1.13 petabits a second – a whole 100 times faster than the first data center network it built for itself a decade ago. Google’s network follows a hierarchical design, with three tiers of switches controlled by commodity chips, which are themselves controlled by software that treats all the switches as one.

Google built its own networks because conventional routers and switches were no longer able to keep up with the company’s needs, said Amin Vahdat, Google Fellow and networking technical lead.

“The amount of bandwidth that we have to deliver to our servers is outpacing even Moore’s Law,” Vahdat said. And in future it will need even more bandwidth to take advantage of faster storage technologies like Flash

“We could not buy, for any price, a data-center network that would meet the requirements of our distributed systems,” Vahdat said. That forced Google to build its own software-defined networks with generic hardware and merchant silicon, the same chips that white-box manufacturers use. Google owns the software stack that controls everything, but it works using the open-source OpenFlow protocol. The company is now onto its fifth-generation homegrown network which it calls Jupiter, using 40 gigabit ethernet connections and three tiers of top-of-rack, aggregation and spine switches.

At the same summit, Microsoft’s CTO Mark Russinovich told the audience that Redmond came across the same problem with Azure. Microsoft’s cloud storage and compute usage doubles every six months, and Azure adds 90,000 new subscribers a month, and this places unprecedented demands on its network, Russinovich said. As a result, Azure relies on a virtualized, partitioned and scale-out network design, delivered through software, in order to keep up.

“You’ve got to have systems that are very flexible and also delivering functionality very quickly,” Russinovich said. “This meant we couldn’t go to the Web and do an Internet search for a scalable cloud controller that supports this kind of functionality. It just didn’t exist.”

So, just like Google, Microsoft was forced to build its own software-defined network, writing all of the code by itself. The company has provided a detailed description of its endeavors here, but essentially its uses virtual networks (which it calls Vnets) built from overlays and Network Functions Virtualization services running as software on commodity servers. Microsoft uses Azure controllers established as a set of interconnected services to partion its Vnets, with each service being partitioned to scale and run protocols on multiple instances for high availability. These controllers are established in regions with between 100,000 and 500,000 hosts, and within these regions there are also smaller, clustered controllers that act as stateless caches for up to 1,000 hosts.

What is surprising its that unlike Google’s network, Microsoft Azure’s SDN doesn’t use any open-source content. Russinovich claims that’s because open-source communities don’t provide the functionality Azure requires.

“As these requirements were hitting us, there was no open source out there able to meet them,” he says. “It’s not an aversion to it; it’s that we haven’t seen open-source out there that really meets our needs, and there’s a switching cost that we have to take into account, which will slow us down.”

Image credit: LoboStudioHamburg via pixabay.com