ONF’s Dan Pitt summarizes and reflects on his experience at the OpenDaylight Summit.
I attended the first OpenDaylight Summit last week (February 4-5) at the Santa Clara Hyatt (where our next Member Workdays will be held). The OpenDaylight (ODL) Project, an undertaking within the Linux Foundation, is developing an open-source SDN controller. Nearly 600 people attended the event. I had been invited to participate in a panel, and decided to stay for almost all of the two days.
Here are my main impressions:
ODL announced its Hydrogen release on the first day. It consists of about 1 million lines of code in 250 bundles for 14 projects (modules) from 150 code contributors. Cisco is still overweight in the committers – that group will have to get more diverse for ODL to maximize its credibility. The leadership – Jim Zemlin and Phil Robb of the Linux Foundation, executive director Neela Jacques, and Technical Steering Committee chair Dave Meyer (CTO and chief scientist at Brocade) – set the tone of openness, inclusion, and the learning value of writing code. As a result the project is starting to feel like a community. But I saw only two people from the user community, so ODL remains a completely vendor-led initiative.
My panel on the first morning was chaired by Neela Jacques and included Guru Parulkar (of Stanford, ONRC, and ON.Lab), Christos Kolias (of Orange, our liaison to ETSI NFV, and the leader of the ONF Wireless & Mobile Discussion Group), and Nick Lippis (of Lippis Enterprises, which also runs the Open Networking User Group). Neela did an excellent job preparing and leading the panel to discuss the roles of standards and open-source software in the context of SDN. I explained why standards at key points (as with IP in the Internet and OpenFlow® in SDN) promote valuable innovation above and below them, with interface abstractions enabling the two innovation domains to move independently, and why software standards are more apt to be de facto than de jure. We shared our respective views on the reasons SDN is important for various users. Much of the audience Q&A was directed at ONF, including questions about the trillion dollars the telecom service providers have invested in their current networks, the challenges to interoperability, and ONF’s experience with northbound interfaces over the last three years. The final question for each panelist asked us what our wish list was for ODL. My answer was twofold: actual modular code components so that they really could be plug and play (allowing, for example, ONOS to slide in), and not to waste time on non-programmatic, non-standard, or single-vendor-controlled southbound interfaces. As it is, ODL actively supports the OpenFlow® protocol but also a bunch of others, ostensibly for compatibility with existing gear, under a “service abstraction layer”. I don’t think the southbound interface (to the forwarding plane) is the place to define services.
The panel on northbound interfaces gave our NBI Working Group chair Sarwar Raza a chance to shine, and he did. Other panelists included Dave Meyer, Colin Dixon (IBM), Pere Monclus (PLUMgrid), Mike Dvorkin (Insieme/Cisco), and moderator Uri Elzur (Intel). Like Neela before him, Uri was well-prepared and the panel explored the purposes and variety of NBIs, the nature of information models, and the role of standards. I thought Sarwar did an excellent job explaining ONF’s contributions to understanding and learning about NBIs, and there was widespread agreement on the value of trying a lot of things in code and seeing what works and meets customer needs. Dave was passionate about this last point.
Our friend Brent Salisbury recently left the IT staff of the University of Kentucky for a job at Red Hat, with whom he has collaborated extensively. He and a colleague gave a hands-on tutorial for setting up an OpenFlow-based controller, which ran on pure OpenFlow® 1.3 and OVSDB with no agents. He described a real-life use case incorporating OpenFlow® 1.0, OpenFlow® 1.3, and OVSDB under OpenStack. Red Hat’s base Linux distribution now includes OpenFlow® 1.0 and 1.3, and later in the hallway his co-presenter actually thanked me (meaning ONF, of course) for OpenFlow® 1.3.
Christos Kolias gave an insightful presentation on ETSI NFV and stated that SDN can significantly enhance NFV (which I have long believed). He also shared Orange’s view of network evolution, including the idea that Telco central offices and points of presence would collapse into a Telco cloud. This reminded me of the remark I heard in Bad Homburg in October that the traditional Telco operation support system (OSS) would disappear with the arrival of SDN and, basically, the Telco cloud.
Marc Cohn (of Ciena and chair of ONF’s Market Education Committee) gave a talk on SDN, ODL, analytics, and carrier-grade networks. I liked that he boiled down SDN into common APIs (NBIs), an open [control] platform (of which ODL is an example), and a standard interface (OpenFlow). His comparison chart between an SDN controller and a Telco management system is one you should see. Line by line he nails the differences.
The final act of the closing session was a concluding panel chaired by Dave Meyer that included Rob Dolin (of Microsoft and ONF’s NBI Working Group), Chris Price (of Ericsson), Ed Warnicke (of Cisco), and Chris Wright (of Red Hat). It was a pretty upbeat conclusion, as I don’t think anyone expected the kind of turnout and sense that transpired. I offered a remark to the panel that ONF invited the participants in ODL to share with us anything they learn that would help us do our jobs, either through their companies if they work for a member or through some other vehicle between ODL and ONF. I also encouraged them to share their experiences when their gear gets in the hands of their customers. I realize that the ODL effort is still in its very early days but it’s always important to keep the end customer in mind.
Much of the program was split into four parallel sessions so no one could attend them all, and a lot of the benefit of a meeting like this is talking with people in the hallways and lounges. This I enjoyed as well, including with a lot of our members. I must say, however, that my most enjoyable conversations were with Dave Meyer. One thing he said really struck me. In the future world of software-based systems (not just open-source efforts), companies and consortia will need to realize that the new tool chains, processes, and cultures they develop are as important as the artifacts they produce. I’m thinking about the implications for ONF.
My main take-away from the OpenDaylight Summit is that there are a lot of people writing actual code to implement SDN, and this is a good thing. ONF’s official position on ODL has been to wait and see how the code turns out, and this remains the case. But the concerns surrounding the way ODL was launched a year ago are gradually giving way to better processes and evidence of broader and more merit-based contribution. So I am seeing ways in which both our organizations can contribute to the success of the open SDN movement. ODL still has some openness to prove, and I hope they will. I hope their modules are modular and they take full advantage of OpenFlow® as the open, standard southbound interface. There seems to be fertile ground for collaboration on northbound interfaces, at least for the use-case and customer set of the ODL contributors. I have not wanted ONF’s contributors to be standards-writers rather than implementers but rather people whose day jobs are building products and services; indeed, we have structured many of our practices with that in mind. But we are not the hackers that they (ODL) are. Also, we have the users, and a very long-term vision for SDN. All told, my week tells me that SDN becomes more real every day.
All the presentations and I hope the videos will be posted soon at www.opendaylight.org.
-Dan Pitt, Executive Director