I would like to kick the afternoon summaries off with a prologue based on my notes and inspiration I took from day 1 of the summit.
It doesnt seem like that long ago, it was hard to really see the utility in futuristic visions of cars and televisions connected to the internet. And not long ago, the U.S. was not a culture of early adopters, or at least outside of the highly left-brained community. But, a funny thing happened on the way to the cloud: computers got personal. Somewhere between the personal multimedia effect of web 2.0, the advancements in highly intuitive natural interfaces, and the advancement in the ease of use of new development tools and expressive languages, the increasingly clear vision of web 3.0 is starting to take shape. While the future is still a bit fuzzy, the notions of mass adoption of internet-connected washing machines doesnt seem so strange anymore. And strangely, we want more. America is rapidly turning into a culture of early adopters, and its not because we are getting more risk-tolerant, we're not. But rather we are becoming early adopters as advancements in intuitive technology and customer-centric design are closing the barriers to early adoption: technology is closing the chasm. And the great thing about the improvements in intuitive technology, is that it allows technology to meet us where we are at.
Recently I was with my wife thinking about how we could get her grandmother on email, contemplating how impossible it would be to send her even a basic PC setup to connect and maintain ... then it hit us, use an iPad. When designed with elegance, new technology has the ability to slip into our lives and be adopted almost without noticing ... until you try to take it away, then you wonder how you ever lived without it. For a few years now, I have noticed how some applications that I have seen have gone from being monolithic beasts to being so elegantly designed that you rarely even need to click the help button to manage the application. And I have wondered, when will this type of experience happen in networking. This is the type of experience I see happening with SDN.
When I first sat down at the very familiar CLI of a certain new SDN controller, and realized with one single, very familiar CLI command, I applied a complete policy to an entire class of devices, globally, and in a way where the policy would dynamically follow the workload. Could this simple CLI command have been ... a new form of automation? While the experience felt familiar and comfortable , I realized it took the knowledge I already had, and propelled it WAY into the future. All of the sudden I was doing things that I never though possible in networking, and it was surprisingly ... easy. Sure its new, and that was a bit scary for me as a CLI centric CCIE with almost no programming experience. But what I didnt realize is how my experience would carry over, and how new learning tools and expressive languages are surprisingly easy, and fun. Many engineers are still unclear of the value of SDN: its all in the experience. Whether it will be the experince of finally being able to have your network operate the way you always wanted it to, or the ability to be at the center of the technology that is propelling your business forward and connecting the world in new and dynamic ways, networking is becoming personal. Its time to stop thinking about what the network is and start thinking about what the network could be, what the network will be. And when you think back at how navigation and seafaring, and how the printing press and how electricity changed the world and contrast with the significance of the internet, it showcases the magnitude of the vision we are building. Networking is to the 21st century what electricity was to the 20th, and I challenge each of us to think about this, think about every aspect of life that has changed, that has improved so drastically as electricity has grown and adapted to every facet of life. And consider how much more impact networking can have on this world, think about all that it can change for good. This is the power of Open networking. It is the power to connect all humans together everywhere, and when we do this, many of the challenges this world faces seem to start solving themselves. It's time to network the world.
Open Networking Foundation Chair Dan Pitt gave an excellent presentation not only raising the bar for what a standards body should be, but which also continued to demonstrate that the ONF represents an entirely new type of organization that, rather than having a focus simply on creating standards, has a much more mature and meaningful focus of creating a vibrant and healthy SDN ecosystem. At the previous ONS event, Pitt gave a great overview of the structure of ONF and how it was designed in an effort to learn from both the strengths and weaknesses of both industry working groups and technology movements. In his presentation Tuesday, Pitt provided a bullish update showcasing the proof points from the efficacy of this exciting new model for community leadership. Pitt's approach reminds me of the new definition of leadership that many world leaders are recognizing is needed to facilitate healthy and vibrant ecosystems. This new definition of leadership is no longer about highly individualistic approaches, but rather more of a shepherding approach where leadership is defined not by being the biggest ego, but by being the one who does whatever is needed for the benefit of the community. This approach is summarized eloquently by MIT Media Lab Director Joi Ito in this brief video. (This is repeated from part 1 of my review, as I inadvertently put my notes for Dan's presentation in the wrong order)
Next, Verizon's Stuart Elby gave an excellent presentation that really showcased the extremely powerful business and technical benefits that Verizon's SDN plan will provide. When thinking of Cloud technology, the data center often steals center stage, and in the data center it is a common goal to be able to move workloads dynamically to avoid congestion and optimize performance, and Elby showcased how Verizon plans to enable similar functionality on the massive scale of their global content delivery network using SDN. Elby highlighted the goals of his project noting that SDN presents new ways for applications to find network services, and new ways for the network to identify and dynamically steer application requests to the best service access point by a large number of flexible, dynamic criteria.
Elby further noted how networks today are not yet optimized for the massive growth in machine-to-machine traffic, and provided some examples of how SDN's can help usher in the hyper-connected era where things like cars and household appliances will be connected to the network.
Extreme Networks V.P. of Technology Shehzad Merchant presented next with a great presentation detailing the tremendous benefits of SDN architecture for everything "From Campus, to Carrier, To Cloud." Merchant used a couple of my favorite examples to showcase the power and promise of SDN, first highlighting how SDN inherently decouples network policy from the physical network and naturally associates policy with the user or resource. Controller-based wireless networks are a great example of how effective this method is at implementing and enforcing network policy. Merchant then noted how SDN has the potential to become the "Android of Networking," and I agree that the future of networking should be about downloading apps and plugins to solve challenges like workload optimization and interoperability, among others.
NEA's Greg Papadopoulos followed, hitting on two major critical themes in a presentation called "Open Networking: Investment for Fun and Profit." And I love this title because it captures that first, as dry and technical as we can get at times, we live in a time where technology is growing to play a very personal and very meaningful role in all of our lives, and when we see growing human impact that technology has, it is such a powerful force for inspiring creativity, and particularly with the tremendous flexibility that SDN tools have and the opportunity we all have to network the world in new and creative ways, bringing the future into reality.
The other main theme is that SDN is not only a fun topic, but at the same time it presents so much opportunity for wealth creation. Papadopoulos continued to hit on the key points around hyper-connectivity, about how so many different devices and technologies still stand to gain so much benefit as they become connected to the network. And with the benefits of this connectivity in mind, it is clear the networking industry has tremendous room for growth.
Next up was a block of presenters led by Princeton's Jen Rexford, and this panel hit on a new theme that speaks to the positive and powerful growth of software defined networking. While the previous ONS focused on defining SDN concepts and building acceptance of SDN, this group of presenters started showcasing the future of what is possible with SDN as they showed that with the new powerful development tools and flexibility created with SDN can lead towards solving many problems both in traditional networking and beyond to solve major problems with the internet and cloud computing.
Dr. Rexford covered a very interesting new project called "Serval," which she described as "Software Defined Service-Centric Networking." This project is tackling the well-known problem that TCP/IP is very host-centric rather than being service-centric. Hosts need to be able to find and access a variety of services, and in distributed service networks there are often multiple locations where a service is offered, cloud networking needs constructs that allow hosts to find services rather than specific IP addresses. Likewise, the network needs a construct to understand which services are available in which locations to optimize host access and service delivery. Rexford's project provides strong justification for the points and methods of service abstraction, and appears to be the most advanced and well-constructed project towards this goal that I am familiar with. Serval proposes the creation of both service ID's and Flow ID's as constructs for a new services layer that would exist between the network and transport layers in the TCP/IP stack. Serval utilizes a service-table as a mechanism for distributing service control logic. More about the project can be found at serval-arch.org.
Cornell's Nate Foster followed with a significant proposal addressing many well-known challenges with network state updates, including particularly vexing problems with race conditions caused by state updates and in-flight packets. Prof. Foster started by highlighting a recent well-publicized Amazon Web Services outage to highlight the significance of problems that can occur during network state changes, noting that "By designing the right software abstractions, we can solve the network update problem once and for all." As a solution Foster proposed a simple and brilliant mechanism to ensure consistent forwarding behavior for a distributed system: per-packet consistency. This elegant solution simply ensures that a packet is treated with the same forwarding policies from ingress to egress within an autonomous domain. Foster has already done significant work to prove the mathematical framework and verify efficacy.
Next, Georgia Tech Professor Nick Feamster presented some immensely valuable and very significant work on new challenges with configuration management in highly dynamic cloud environments. Dr. Feamster first highlighted that the tools used today (such as Modular QOS CLI and ACL's) to implement network policy are not very expressive for translating business rules into actionable policy. As examples he cited time-of-day-based restrictions, bandwidth caps, and infected hosts with needs for quarantine and remediation as examples of criteria that is either difficult, non-standard, or impossible to enforce in traditional networking systems. Dr. Feamster expanded the problem definition by noting the extent that network configuration errors still cause network outages, and highlighted how his solution can address this significant challenge. As a solution, Dr. Feamster's team is working on a project called Lithium which utilizes event-based network control. Lithium treats network policies as event-based programs, allowing the controller's finite-state-machine to invoke events to trigger network state changes. One example that Dr. Feamster highlighted was a new approach to implementing NAC within SDN architecture which in my view could offer a significant improvement over the solutions available today.
Rounding out the day, Prof. Jun Bi of Tsinghua University gave an excellent presentation on remaining challenges for SDN and OpenFlow deployment in large-scale networks. Prof. Bi covered so much research in his presentation it is hard to summarize, providing a great showcase of significant contributions to networking research. First, Prof. Bi showed some excellent work done on adapting and testing OpenFlow in IPv6 environments. Next, he noted challenges with uRPF amplified by IPv6 as well as extensive multi-pathing and flow steering in software defined networks. Prof. Bi reported on a solution called "Calculated Path Filtering," which is a mechanism that takes advantage of the SDN global control plane to create an interesting test for the evolution of uRPF. Prof. Bi's model leveraged an innovative approach to accomplish the goal leveraging some proposed OpenFlow tlv's and distributed routers. I am not convinced that this model of distributed routing will continue unchanged as SDN architectures advance, but it is an interesting showcase of a hybrid approach which could provide an example of how existing NOS control-plane functions could be repurposed. I will be interested to see how this would function with a routing model like RouteFlow, or other approaches that utilize centralized L3 control plane services distributed via intelligent agents.
Prof. Bi continued reporting on research on a model for named data networking, which has a similar goal as Serval: to change the internet routing model from being host centric to being service centric. He proposed an interesting model that would include 3 tables in the network forwarding plan, a "content store", a "pending interest table" and finally the standard FIB, providing an interesting exercise in considering the implication of a post-IP, service centric network. Finally Prof. Bi gave updates on projects advancing NOS challenges through a virtualization platform, and also on a framework for dynamic negotiation of Inter-domain policy negotiation.
To close I would like to borrow a slide from Dan Pitt's excellent presentation that I think captured the spirit of the day: