[Editor’s note: Curt is an extended member of the Wiretap/SDxCentral family and graciously agreed to help us cover ONS while Matt and Roy were fulfilling multiple duties at ONS 2013 on Wednesday. If you haven’t read the highlights from Tuesday yet, check it out first, along with Wednesday Part 1. ]
Plenary Session: SDN for Service Providers, Part 2
“SDN@Google: Why and how” by Amin Vahdat (Distinguished Engineer, member of the ONF Technical Advisory Group at Google)
Amin started his talk with, “If Google were an ISP, it would rank as the 2nd largest carrier on the planet,” and a graph showing how its public network accounts for ~6% of all internet traffic! He added that their private network is larger and growing even faster… But due to WAN cost problems, Amin said that they could not use existing infrastructure (similar to the Telstra session). They are using an SDN WAN to separate hardware from software, logically centralize network control, and automate everything. Google runs their own network hardware based on merchant silicon (accessed via OpenFlow) and uses Quagga for BGP & ISIS routing.
Google migrated functionality from switches into their controllers over time, and now have one datacenter’s WAN fully under SDN control. Their traffic engineering server understands network topology, knows about failures, and installs routes that override the default shortest-path forwarding (but can fall back if needed). Their WAN links run at almost 100% utilization, and high priority traffic doesn’t experience any loss. Amin concluded that the implications for ISPs are clear: SDN can dramatically reduce WAN costs, act as a differentiator, and be deployed incrementally in a pre-existing network. In short, it’s real and it works. Overall an interesting presentation and a vivid account of how to tackle problems the Google way.
“SDN: Time to Accelerate the Pace” by Justin Dustzadeh (CTO & VP of Technology Strategy at Huawei Networks)
Justin opened his presentation with a clever application of the “5 Stages of Grief” to SDN. He cited phrases and quotes that one would hear from people at each stage (denial, anger, bargaining, depression, and acceptance). It was a funny and lighthearted way to push back at negative critics. After a brief introduction to Huawei (wah-way – not Hah-wah-ee, like the island), he proceeded to talk about SDN principles and why customers should bet on their carrier-grade approach.
To that end, Justin spent the bulk of his presentation going through 4 use cases that were wide-ranging and touched every type of carrier network. These spanned residential optical access (Telefonica), IP RAN (China Telecom), next-gen POPs (France Telecom), and cloud data center interconnects (China Unicom). All of the examples lacked hard data on benefits, but this is understandable since the trials either just started or will do so by the end of the year. The latter parts felt like a company & product pitch, except for the final topic, “protocol oblivious forwarding.” In the end, Justin made a fairly effective case that Huawei has fully embraced SDN for carriers.
Plenary Session: SDN for Enterprises & Enterprise Data Centers
“Protocol Independence” by Nick McKeown (Professor of Electrical Engineering & Computer Science and Faculty Director at Stanford University, Board Member of ONF)
Nick started his spirited and fast-paced presentation with a brief discussion of SDN, its technical consequences, and some history. He also introduced the notion that the original goal of OpenFlow was protocol independent forwarding controlled by a separate control plane. And unfortunately, the current incarnation of OF remains protocol dependent, forced into existing silicon for time to market reasons and ultimately suffering from those constraints. Nick then asked the audience “What would an OF-optimized switching chip look like?” and proceeded to lay out the high-level chip requirements and its design from top to bottom. I’ll spare you the technical deep dive here, but highly recommend getting a copy of the slides if / when they appear on the ONS web site.
Nick did this design exercise with Texas Instruments, and they concluded that it would need < 15% additional area and power versus comparable chips. He used the results to conclude that OF can and should evolve into something much more versatile than it is today. Whether you agree with Nick or not, he is without a doubt a great presenter and fearlessly dove into a blistering technical onslaught with ease. It’s possible that many in the audience were completely dumbfounded by the entire thing. It was the last session of the day, so I wasn’t able to gather reactions from attendees. But as recovering device driver software engineer, I loved every minute of it. It felt like Nick was throwing a gauntlet down to vendors and saying, “What are you guys doing? Why haven’t you made a chip like this?” Whether you liked it or not, it was nice to see some content that really forced the audience to think, presented by one of the preeminent visionaries in the SDN world.
Final thoughts and wrap-up
So with a review of the sessions behind us, here’s a quick look at a few of the key take-aways from this year’s event:
- The SDN world is evolving, although not as quickly as in the past (which has pros and cons).
- Enterprise customers and Service Providers are showing up with rich, real-world SDN use cases deployed in production networks that they can talk about.
- These same deployments appear to be light on hard data about benefits and ROI.
- The SDN world is divided into at least 2 distinct camps
- Large-scale public cloud (IaaS) providers and vendors who use overlays to get flexible network virtualization
- Pure-play OpenFlow networks that are more circuit-like in nature and appeal to carriers that need simple and automated provisioning.
- The market for SDN silicon is an interesting and turbulent place to be.
- There is a lot of interesting work being done on SDN at research universities.
I definitely enjoyed my time at ONS 2013 and the gracious hospitality of our hosts, Guru and Sedef. To wrap up my coverage, I’ll list what I heard from other attendees on what ONS can do to make next year’s summit even better:
- The session content can improve (too many company- and product- oriented pitches)
- Hold more technically-oriented sessions, with adequate time to get deeper into the material
- Vendors need to present their material and ideas in better and more creative ways
- Vendors should put more customers on stage time and let them relate their vivid experiences because this is the stuff that sticks in people’s minds
- We need to see more hard data on benefits and measured ROI
- More news keeps it more interesting (announcements, technical approaches, etc.)
- The research track could have fewer sessions (best of the best) with a more time for each
- Adding an hour to the 2nd day would give more time for exhibits after lunch and a badly needed break between the afternoon plenary sessions.
And finally… although I didn’t attend the developer track, the unanimous feedback I heard was that it needed to provide much more technically-oriented information. Engineers tend to have very little patience for marketing or business “fluff”!
If you have thoughts or comments, please feel free to contact me or the Wiretap/SDxCentral team. Like me, we hope you had a good time at the show and look forward to seeing you at the next one!