OpenFlow (and other network heresy)
kgray

Welcome to heresy, welcome to OpenFlow, and welcome to Japan (Tokyo InterOp)

by Ken Gray (kgray) on ‎07-16-2012 10:19 AM

"A Heresy is any belief or theory that is strongly at variance with established beliefs or customs."


It's been an interesting year here at Juniper and an interesting year in our industry. 

 

As this year's follow-on to the "cloud" phenomena, SDN (Software Defined Networking to some, Software Driven Networking here at Juniper) is creating more excitement and shows more potential (if the buzz is to be believed) than anything in a long time (in networking). 

 

At the vanguard of the SDN wave, is the Open Networking Forum (ONF) and it's OpenFlow protocol ..the first protocol with it's own Marketing firm :smileyhappy:.  While I would argue that OpenFlow by itself is not SDN, the two concepts are linked, directly by the marketing of the ONF and by inference in their evolution as concepts.

 

Because OpenFlow advocacy often pits a centralized control plane approach to networking against the more traditional ideas we developed as a company to lead the industry to where it is today ...a distributed control plane (particularly MPLS and BGP), I am unofficially the heretic in our internal, professional and collegial discussions of networking here at Juniper.  I am one of a small group that represent Juniper at the ONF.

 

I embrace the role because it is an outward expression of Juniper's commitment to pursue "outside the box thinking", innovation where it may potentially arise, and because some of the concepts found in OpenFlow (and subsequently SDN) continue the theme of Network Programmability that Juniper began sharing with customers a couple of years ago. 

 

In discussions with colleagues, I find myself often referring to customer perception (based on discussions and participation in customer events).  With that in mind, no matter what you may think personally/professionally of OpenFlow and SDN, if these discussions and the level of participation in industry events (by vendors and customers alike) is any indicator ...there may be something of substance in OpenFlow and certainly in SDN.  These concepts are still fairly young, the OpenFlow protocol is admittedly still a bit brittle and much of what was exhibited at InterOp had the air of "trial" (vs. "hardened"), but the level of customer interest has been really impressive and worth some consideration.

 

I attended both InterOp Vegas and InterOp Tokyo as a speaker and booth worker (focusing on the OpenFlow InterOp booth in Las Vegas and the Juniper and partner booths in Tokyo).  While there was a LOT of OpenFlow/SDN at InterOp Vegas, Tokyo made that level seem trivial.  It could have just been the momentum built up in the two intervening months or that Japan is a much more aggressive environment for these topics.

The SDN/OpenFlow private seminar we (Juniper) gave in Tokyo (prior to show) had as many attendees ( a very full room) as some of the public panel sessions I participated in at Las Vegas InterOp.  My colleagues (Hidet Sugiyama, Nils Swart) and I all took 40 minute segments speaking on various aspects of Network Programmability, SDN/OpenFlow standardization and practical deployment.

Unlike Las Vegas, where Juniper only had an interoperability demo (with Big Switch Networks), Juniper had several notable demonstrations on the InterOp Tokyo show floor, including; the EX4200 VC being used in ShowNet as part of the Coordinated Threat Mitigation section of the network (security controlled by OpenFlow controller app), the Bandwidth Calendaring Application (our SDN framework example, using both PCE and OpenFlow for path provisioning) in partner Nissho's booth and an inter-DC connectivity demo using OpenFlow into lt-lsys-pw/lsp (done in cooperation with NTT) in the OpenFlow Showcase booth.

 

There were few, if any, booths at Tokyo InterOp that did not have a mention of OpenFlow or SDN.  Some booths seemed completely OpenFlow/SDN focused (Intel/HP, NEC for example).  The OpenFlow Showcase booth was very crowded every time I passed and the seminar space (OpenFlow seminar every half hour) was standing-room-only continuously every afternoon (I have pictures of different seminars with over 200 people in attendance each time ... here's one).

 

IMG_0656.JPG

Michael Beesley's (our Platform Systems Division CTO) keynote on Network Programmability was also SRO.  On the day he spoke, a salesforce.com booth tracking app (tracking salesforce posts and other social media mentions of InterOp Tokyo) showed the most used phrases were "SDN" and "OpenFlow" in messaging for the day.

 

What is hard, at this point, is to separate how much of that enthusiasm was for OpenFlow-the-protocol and how much is for the concept of SDN (because the two topics are often conflated).


The observations that OpenFlow and SDN are (at least) popular, are not necessarily heresy for the traditional networker.  What will make them seem more so is the cross-pollination and extension of OpenFlow with some concepts particular to routing and optical transport (Layer 1 and Layer 3 networking ... which is already starting to happen today under the banner of SDN via ONF's Hybrid Working Group or Extensibility Working Group and through drafts appearing in the IETF) and the re-evaluation of almost every networking "solution" in light of whether it can be achieved more efficiently, cheaply or simply via OpenFlow and/or SDN (through centralized control, distributed control or some hybrid).

 

For examples of this sort of hybrid thinking, check out http://www.ietf.org/id/draft-medved-pwe3-of-config-00.txt  and https://www.opennetworking.org/bugs/browse/EXT-154 (the latter may require you to register with the ONF Jira site) for perspectives on how pseudowires may be mapped to MPLS-TP tunnels and optical transport networks manipulated using OpenFlow.

 

What we have to do as an industry (and what I hope to treat in this blog ...along with other heretical thoughts) is tease apart the SDN and OpenFlow relationship, further defining the purpose and roles of both and where they intersect, overlap and compliment traditional networking. 

 

While you're waiting for my next post, if you need an SDN fix, I suggest you read the blogs of my colleague Thomas Nadeau (Occupy SDN).

 

 

Copyright© 1999-2013 Juniper Networks, Inc. All rights reserved.