Ananda Rajagopal is vice president of product management in the Data Center Networking Business Unit at Brocade, responsible for the routing portfolio used by software-defined data centers, wide area network providers, and large enterprises. He has been associated with several seminal developments in our industry, including one of the first large-scale SDN implementations on a 100-Gb/s WAN, the industry’s first multi-terabit router, the first large-scale IPTV deployment, and many others. Ananda joined Brocade through the Foundry Networks acquisition. Prior to Foundry, he served in engineering management roles at Lucent and Ascend Communications; Cisco and Stratacom; and Wipro.
Mike Capuano joined Infinera in May 2011. As vice president of corporate marketing, he is focused on strengthening the Infinera brand and increasing awareness of its solutions and value proposition as the company expands its product portfolio and global reach. With his extensive background in routing and transport, he is a key leader for Infinera’s software-defined networking (SDN) initiative and speaks frequently on the topic. Mr. Capuano brings more than 20 years of experience in marketing in the technology and networking industries, including executive marketing positions at Cisco Systems and Juniper Networks.
With regard to the ESnet multi-layer SDN demonstration, what was Brocade’s primary role?
What’s unique about the MLXe that made this demonstration possible?
Rajagopal: It is a combination of things: the ability to handle very large flow sizes (up to 100 Gb/s), full support for OpenFlow without any performance impact, high rate of flow setup handling, and the ability to do all of these for many tens of thousands of flows.
Was SDN necessary to make this happen? Why?
Rajagopal: To date, there has been a lack of common management abstractions that work across multiple layers in the network in a multi-vendor environment in a simple way. Achieving multi-layer network control that spans both the packet and optical layers in a network between, say, two data centers has been impossible at worst, or far too complex at best, to achieve, especially in a multi-vendor environment. Customers are yearning to create open architectures that facilitate multi-vendor, standards-based, best-of-breed solutions, and this joint demo is a proof point of how that can be realized with SDN. The picture below captures the problem today and how SDN can be used as a common abstraction to drive such network control, resulting in true operational simplicity.
What do you see as the key lessons learned from the ESnet demonstration?
Rajagopal: SDN today is no longer theoretical; practical use cases have emerged, and customers for the first time are empowered with tools to drive multi-layer networking efficiency in the network — with little hand-holding from vendors and no changes in vendor software. When we were designing OpenFlow on our products, we did not target this specific use case. That is the benefit of abstractions; by viewing OpenFlow as an API to the data plane, the speed of change can be accelerated by the customer even for use cases unanticipated by a vendor. To draw a parallel, can you predict the next app that will come out on your smartphone’s OS? No. SDN has allowed networking to begin that same evolution.
How do you see this translating into benefits for Brocade’s enterprise and service provider customers?
Rajagopal: Brocade’s customers have been deploying SDN-enabled routers for over seven years! Of course, SDN did not even exist in the industry’s lexicon for much of this time, but it speaks to the benefits of the inherently programmable and future-proof architecture on these routers. We have deployed over 1.1 million router ports over this period, every one of which can run OpenFlow SDN with a software upgrade. That is powerful, because it debunks the notion that SDN requires a forklift. Coupled with the hybrid-port-mode capability offered on the MLXe routers, enterprises and service providers can begin the SDN journey by starting with specific use cases and with minimal impact to their present mode of operation. The ESNet-Infinera-Brocade joint demo also shows that customization specific to the participant vendors’ products is not required, creating a truly best-of-breed, multi-vendor and open environment with out-of-the-box hardware and software. That is the power of SDN.
What was Infinera’s role in the ESnet multi-layer SDN demonstration?
Capuano: SDN has historically been focused on layers 2, 3, and 4, and primarily inside the data center; with this ESnet demonstration, Infinera has taken a leadership role in expanding SDN down to layers 0 and 1. Specifically, we developed a lightweight virtual switch called the Open Transport Switch (OTS) that runs as a software plug-in on our intelligent DTN and DTN-X optical transport platforms. OTS presents a virtual transport topology and enables topology discovery, configuration, monitoring and provisioning of the virtual transport network. For example, from a configuration perspective, OTS maps the virtual transport representation to actual physical resources in the transport layer.
Can you share more about the demonstration and the key benefits?
Capuano: There were two parts to this demonstration. The first was on multi-layer provisioning, and the latter on multi-layer optimization.
Long-haul optical transport is used to connect routers together over long distances, and the “router layer” and the “transport layer” are typically run by different groups and use different provisioning and management systems. So, assume a new service is sold that has to be delivered by routers, and that service requires additional transport bandwidth between two cities. With the typical manual processes between these two layers, it can take a few days or even weeks to get that capacity provisioned. The first phase of this demo showed that if there is a virtualized intelligent transport network that can be provisioned by OpenFlow working in conjunction with an OpenFlow-enabled router, then both layers can be provisioned simultaneously by an SDN controller. This offers not only significant operational cost savings but also speeds service delivery, enabling service providers to be more competitive.
For this to work, of course, you need a virtualized, intelligent transport network. Historically, optical transport has consisted of dumb point-to-point pipes based on transponders. This demonstration shows that an intelligent optical network with converged OTN switching and DWDM in a single platform allows the creation of flexible mesh networks with the ability to provision virtualized bandwidth with granularity from 1 Gb/s to 100 Gb/s. So, an important part of this demonstration is to show that OTS working with an intelligent transport network can rapidly allocate bandwidth services in a dynamic manner.
The second part of the demonstration shows how multi-layer optimization can be triggered by specific events. In this situation, the demonstration showed a small flow growing into a larger flow could trigger the SDN controller to take an action to optimize the network by bypassing an intermediate router. By leveraging the intelligent transport network and SDN controller, the transit traffic from the midpoint router was removed. This saves capital expenditure by reducing the need to buy more ports on the midpoint router, saves power by not using excess equipment, and the automated nature of the optimization saves operational overhead and time.
Was SDN necessary to achieve the results in the demonstration?
Capuano: Technically, SDN would not be required to execute this type of demo, but it makes it a lot easier, so you could consider SDN a very good approach to achieving this. Routers today tend to only see the network through adjacencies, while SDN has a single global view of both layers of the network and can make networkwide decisions. This, coupled with standardized southbound interfaces to a large array of multi-vendor equipment, allows for a distinct advantage over previous approaches to solving this problem.
How do you see SDN affecting GMPLS? Does it replace the investment Infinera has in GMPLS?
Capuano: Not at all. GMPLS will remain a critical intelligence element of our solution, and I see them as complementary. With SDN, we have two different modes, implicit and explicit provisioning. In implicit provisioning, SDN is able to use a centralized view but can initiate a request from a single node, which in turn uses GMPLS to set up an end-to-end path. This probably fits with what many service providers are comfortable with today and provides a nice migration path. If there are service providers that want more direct control, we have explicit provisioning, where the SDN controller performs direct programming of all the nodes. However, GMPLS is still used to discover and track topology and can play an important role for providing protection and restoration capabilities for transport flows, which need localized switching speed. So in all cases there is a critical role for GMPLS.
What are your key takeways from this demo? Is SDN ready for production now?
Capuano: We certainly have made lots of progress. We did a demo last year with ESnet on just simple provisioning at the transport layer only, using an early version of OTS, and now we are in a multi-layer provisioning and optimization demonstration with a much more advanced version of OTS. The additional enhancements we’ve made to OTS recently include a RESTful/JSON interface which now provides topology exchange and configuration in addition to monitoring. Definitely steps needed for service providers to start looking into deploying this solution.
One key takeaway for me beyond the value of multi-layer provisioning and optimization is the fact that Brocade and Infinera were able to work with an open controller and show that the service provider ESnet was able to write their own application to control the network. The idea of SDN is an open ecosystem where the openness will drive innovation from any participant. SDN will drive innovation, but not if vendors are deploying closed systems and calling them SDN. For example, some vendors in the transport space are proposing APIs into their existing EMS systems. Well, that means that any innovation that a service provider might want in a multi-vendor environment is going to have to wait for the release cycles of very large, heavyweight EMS solutions — and that just does not seem all that innovative to me. With the lightweight OTS approach, changes can be effected quickly and foster rapid innovation — which is the fundamental principle of SDN.
With regards to progress, SDN in the WAN is still in its embryonic stage, and there is lots of work to do to make it carrier-grade — scalable, reliable and secure. I see 2014 as a year of continued development for the entire industry, and testing with some production pilot announcements, but with more meaningful production rollouts to happen in the 2015 timeframe and beyond. We’ve come a long way, but there’s still a lot of work to be done.