OpenFlow Symposium in San Jose – Networks as Application Stacks
on 10-27-201110:41 PM - last edited on 10-28-201308:50 PM by bcm1
On October 26, an OpenFlow Symposium sponsored by Packet Pushers and Gestalt Networks TechFieldDay was held at the DoubleTree hotel in San Jose. It was a day to listen to the vendors working with an emerging architecture, called Software Defined Networking (SDN), and a research project, OpenFlow, that is out of the lab and being actively developed by the Open Networking Foundation into a programmatic interface.
Moderators Ivan Pepelnajk and Greg Ferro
So, what’s going on?
There are various descriptions of SDN and some over-reaching and hype about OpenFlow. Scott Schenker gave a talk on the architectural limitations of networking that slow its adaptation. He points out that computing and applications have developed a more efficient programming architecture which is why applications can change much more quickly than the underlying networks they rely on. Since networks exist to transport application traffic, if the network can’t keep up with the changes in applications, the network architecture becomes ripe for “adjustment”. Enter SDN and OpenFlow.
Scott explains how the architecture of applications and computing relies on abstraction to simplify the complex via modularization, and how programmatic interfaces simplify the synthesis of modules into complex application stacks. This architecture “extracts simplicity from complexity” while networking today relies on “mastering the complex”, creating two very different mind sets and cultures. The OpenFlow and ONF effort are creating a different architecture based on abstractions that simplify the complex using a programmatic interface, OpenFlow, so networks can evolve at a comparable rate to the application stacks.
The network vendors are engaged with this transformation centered around the SDN architecture and OpenFlow as one way to create a programmatic interface for the control plane of the network. Big Switch , Brocade, Cisco, Juniper, NEC, and others including start-ups, are contributing to work on this new network “as an application stack”.
Here is a summary of yesterday’s OpenFlow Symposium, which was broken into a morning, business perspective and an afternoon technical perspective.
Symposium Live Attendees + Several Hundred via Live Web Cast
This session was expected to focus on use cases for OpenFlow, with emphasis as much as possible on real-world experiences.
First speakers were from Google & Yahoo, who were rather vague about what they’re actually doing with OpenFlow, to the disappointment of many. Big Switch followed.
Brocade provided a practical reality check on what OpenFlow is and isn’t, and next steps to take if interested. In retrospect, many felt Brocade should have gone first to help set the stage. The Cisco presentation segued nicely from Brocade’s intro into the details of the technical challenges OpenFlow attempts to solve. There were some questions on OpenFlow’s true scale from the crowd.
NEC, with OpenFlow-enabled products in-market, presented on two customers using OF. Left the audience pleasantly surprised at how close to “reality” OpenFlow might be.
Juniper focused on using OpenFlow to better support app needs. Some of the justification seemed to be that VC money is going to apps, and OpenFlow will help the NWing industry jump on that gravy train again. Juniper also made an announcement of OF support on the JunOS SDK yesterday.
Videos of all of the presentations, as well as most of the slides, can be found here.
It was clear that several of the attendees were eager to understand the likely rate of adoption of OpenFlow and how serious the vendor community really is about making it a reality. The panelists, as well as a representative of the ONF, took pains to stress that while some very early adopters are already experimenting with OpenFlow 1.0, the technology is still very nascent, with many questions still to be hashed out. “We’re still learning how to do this” was a common remark.
40-50 live, highly focused attendees with many more on the live internet feed. Unfortunately, the feed went down for about 45 minutes due to network failure … the irony was not lost on the attendees.
This session got quite dense very quickly and was fairly wide-ranging in nature. Topics included:
What is Software-Define Networking, exactly? (General consensus: what you get when you properly orchestrate the full ecosystem of NW, virtualization and controller vendors to get OF working)
What forwarding functions should be included in OpenFlow? Basics, or kitchen sink?
Appropriate controller scope: distributed, hierarchical, by activity, etc.
How OpenFlow will affect networking vendors, and potentially reshape the vendor landscape
Many more…Ethan Banks provided further details in his writeup shortly after the event.
The latter portion of the afternoon was dominated by frequently sharp back and forth between Juniper and the Yahoo-Google axis; other vendors stepped in from time to time to add color and nuance.
Most attendees felt that the afternoon was the most interesting part of the day. (I certainly learned a lot!) There was a lot of praise for panel’s collective technical expertise, seriousness and lack of hype. (Audience quote: “I haven’t seen a unicorn all day.”) A couple of other attendees we spoke withafterwards agreed that the discussion made them realize how embryonic the whole thing is, but also how quickly things are moving.