Google Showcases OpenFlow network

Google kicked off a great day of presentations at ONS 2012 Tuesday by highlighting their journey into openflow and showing that their massive inter-dc network, one of the largest networks in the world, is already running on a 100% openflow based network.

Google kicked off a great day of presentations at Open Networking Summit (ONS) 2012 Tuesday by highlighting their journey into openflow and showing that their massive inter-dc network, one of the largest networks in the world,  is already running on a 100% openflow based network.

RELATED: Google shares lessons learned as early software-defined network adopter

MORE: Open Networking Summit 2012: Google, Verizon, NEC, others tackle future of OpenFlow

Note: This report was not prepared, reviewed or endorsed by Google, images used below are from Google's presentation at ONS 2012 and will be publically available at the ONS Website.

Historically, OpenFlow has faced some tough opposition from the old guard of the networking industry, despite some more recent support which has not yielded much of anything in terms of concrete plans and still yields continued emphasis on proprietary alternatives. I have long felt like there has been an heir of intellectual snobbery from the legacy networking titans, acting as though the thought that someone other than them could actually come up with a network innovation was laughable, as though networking was much more complex than these academics and cloud dreamers realized.

So I imagine it must have felt pretty damn good to stand up and tell the story of the near seamless migration of one of the world's largest networks to OpenFlow. A story that now cant be simply dismissed as there are now facts on hand that can no longer be argued with.

Google SVP Urs Hölzle was the storyteller, and the backdrop for the story was Google's efforts to drive low cost, flexible services to power cloud economics, as Hölzle noted, Google's services are often free, or if not extremely low cost … so decreasing the cost of the network component  relates directly to the availability and quality of free and low cost services that many of us use every day. In an enterprise context, similar network cost benefits will become increasingly crucial as cost pressures from outsourcing providers continue to intensify.

In other domains of computing, economies of scale have a significant impact. Hölzle noted that in applications like enterprise email, Google  has been able to achieve ~100x improvement in efficiencies from economies of scale. However, this is not the case for traditional networking solutions which do not receive any notable benefit in cost per bits as the backbone scaled. The sources of the high costs included the high cost of large routers with MPLS capabilities, the cost and complexity of managing traffic engineering, trying to manage, let alone orchestrate  network systems and challenges with the overall architecture of traditional solutions.

The main benefits that Google claims thus far primarily relate to the concept of managing a single cohesive system versus managing a bunch of independent autonomous devices. Whether trying to accomplish something that should be simple like a basic resource reservation or something more complex like MPLS-TE, traditional networking approaches … suck (Why should a simple desire to do basic optimized routing require something as complex as MPLS-TE?). Hölzle went through some simple examples of failure and convergence scenarios with traditional approaches showing that failures were non-deterministic, difficult to manage and troubleshoot.

In contrast, SDN (Software Defined Network/ing) provides a global view of network utilization, allowing simple and dynamic traffic-steering on low cost hardware. SDN provides deterministic behavior which not only can create superior customer experience and better SLA's but also can address  the historical problem of having to overprovision network capacity. Hölzle further noted that the control-plane for most routers today use substantially worse hardware than what is available on server platforms, noting that modern server platforms can achieve 50x greater performance.

One of the strongest benefits with SDN for Google has been the robust testing capabilities available with SDN. Hölzle reported that the transition from traditional to OpenFlow networks was nearly seamless and they encountered far fewer challenges than expected. Much of what made this possible was the ability to very easily simulate backbone-scale network environments virtually, including the ability to mirror production event streams in testing environments, which allowed Google to identify and fix all the bugs in advance.

Perhaps the coolest  part was, during the Q&A an audience member asked how Google was able to get their NOC up to speed quickly with the new network, and the response was that, the robust simulation and virtualization capabilities made it easy to train noc staff on real-world simulated networks before they were ever put into production. And because the test environments can provide accurate topologies using real binaries and real production traffic, training the NOC staff was a piece of cake.

The process to transition from the old to the new network was essentially to bring the new network up, at which point the old network created new adjacencies and discovered what appeared to the legacy network simply as new routers, allowing a  graceful migration as they simply started steering flows over to the new network.

What does this mean for enterprises today?

The first thing is that, it does not mean you should go try to deploy an openflow based SDN on your own. The Open SDN movement behind OpenFlow is often considered to be analogous to the server industries movement from proprietary vertically integrated mainframe model to the open x86 architecture. If you consider this analogy, it highlights that openflow is only one piece of several that are required to deploy a software defined network. Google employs a massive staff which includes many of the top computer scientists in the world, and their business model allowed them to develop a comprehensive suite of hardware and software  to support this deployment.

What it should demonstrate however is that the technology and architecture work, and provides substantial benefits. And while Google has a lot more scale than the average enterprise, I think you'd agree the benefits that Google has found with SDN would be equally compelling for companies of all sizes, and as Google indicated in their presentation, this is just the beginning.

Enterprise networking vendors today are just getting off the ground with OpenFlow and SDN solutions, but, it’s the cloud era, business competition is increasing, and pressures to use the best possible technology quickly are mounting. While it requires some boldness, as in Google's case, a well planned early adoption of new standards-based technologies can give the adopting business significant compatetive advantage. The capabilities adn benefits of OpenFlow are very significant, think this will be in demand within enterprises sooner than some might expect.

Another thing I have observed is that SDN can also be rolled out in a way that utilizes very familiar tools and processes that requires very little training for most network engineers today. This will vary significantly based on the vendor, but good SDN solututions will have a comprehensive solution for graceful migration for the rekevant segments of a network, at any pace. The vast benefits of SDN Architectures could propel SDN into enterprises very quickly, the technology is mature, and at least some enterprise networking vendors already have OpenFlow solutions available, others will have SDN solutions available very soon. 

But as OpenFlow 1.0 Author Guido Appenzeller noted in his presentation, the biggest threat to SDN today is proprietary networking alternatives. I believe Guido was referring to proprietary SDN's built using openflow or similar, but I think this also applies to a lot of pre-SDN 'next-generation' proprietary architectures. The old guard of network vendors dumped very large sums of money in these architectures, and I think it will be interesting to see if they can create a graceful migration strategy that leads towards the multitude of benefits that can be accomplished with SDN.

In any case, organizations that is invest in network technologies today should weigh heavily the vendors approach and migration path to SDN. A good first step would be to talk to some knowledgable resources about how SDN architecture can solve your specific business challenges and provide benefit to your environment. If SDN architecture looks like it would have benefit to your orginizagtions, investigate further. Dig in to your vendors architectrure and roadmap and find out the details of the migration path between what they are selling today and how it leads towards the tremendous benefits of SDN. Also, watch out for proprietary SDN approaches, a lot of networking vendors fear that openflow will commoditize networking gear, which would be great for consumers, but bad for companies that think they need a proprietary approach to lock you in to their business model. I think we all know that following periods of great change, there are periods of consolidation, and with a little time the networking industry will be due for some Betamax wars where a lot of the proprietary and one-off architectural approaches will get flushed. If I wanted to make a bet today for which architecture will be dominant or even still around in 5 years, compared to any other 'next-gen' architecture, OpenFlow SDN architecture is the only one with broad industry consensus, and if we are headed for a Betamax war, I would like to be on the side of pretty much every single one of the worlds largest cloud providers, service providers  and network operators that have thrown their weight behind openflow.

If you have read my column, you know I haven't been big on product plugs, but I am a Dell Employee. I am not writing this blog as a Dell representative, anything I say here is my opinion, and I would like to share that not every networking vendor bought into the old way of doing things. So I know at least one vendor has been using the newer style of low-margin, high-performance networking hardware for some time, and has a robust and smooth migration path for the solutions we sell today into a robust standards-focused SDN architecture. Use that to keep other vendors honest, Cloud providers and the ONF are fighting for a more open, healthy and standards focused industry, this is the vision of what the internet is supposed to be, but it will only happen if each of us demand it.

The final conclusions of Mr. Hölzle's presentation:

Copyright © 2012 IDG Communications, Inc.

The 10 most powerful companies in enterprise networking 2022