What is Avaya's Strategy Around SDN? A Q&A with Avaya Networking's Chief Architect, Paul Unbehagen
With all of the recent interest in Software-Defined Networking (SDN), the networking space suddenly feels as hot as it was during the late 1990s. Most vendors are aligning themselves into one of two camps, either throwing their public support behind the OpenFlow communications protocol or jumping onto the OpenDaylight open-source bandwagon.
At Avaya, we don’t think the choice has to be so black-and-white. I recently interviewed Avaya Networking’s Chief Architect Paul Unbehagen.
Unbehagen has 24 networking patents and is co-author of the IEEE 802.1aq specification for the next-gen networking technology, Shortest Path Bridging. He was joined by another Avaya expert, Rob Turner.
What’s the biggest myth being spread about SDN?
Unbehagen: People are talking about SDN as if it were a new thing. It may be in the news, but it’s not new. In fact, the concept of separating and abstracting the different elements of network in order to manage it more easily and more powerfully – that’s been around for at least 20 years.
Take Asynchronous Transfer Mode, or ATM. Every time you create a Permanent Virtual Circuit (PVC) in ATM, you’re asking a controller to automatically build a path from one location to another by enabling it on every switch that the packets travel through. That’s the same concept as today’s SDN. Every optical network around today also uses a form of SDN. So do wireless LANs.
When you take a step back, you see that SDN, perhaps under another name, is something the industry has been doing for many, many years. Now, we’re applying it in new ways. Ultimately, true automation and virtualization will come from marrying the abilities of a network controller with the fabric. This will provide a whole new level of flexibility in data center design and operations.
Related article: Top 10 Networking Myths
What about OpenFlow?
Turner: A lot of people are equating OpenFlow with SDN, when the latter is a broad umbrella concept. It’s like calling all smartphones ‘iPhone’ devices, or all soda drinks ‘Coke.’ Some proponents are also championing OpenFlow as the great solution for the enterprise. It isn’t.
OpenFlow’s claim to fame today is that it gives network admins a programming interface so they can reprogram and override what their routers and switches do today. While all enterprises have programmers, rarely do they employ network programmers. Most never have and never will. Also, your router already knows how to optimize network traffic paths better than a network admin ever will.
Most importantly, enterprises want solutions that automate network management, not add complexity. OpenFlow’s programming interface just gives enterprises one more potential problem to solve.
So what is Avaya doing in SDN today?
Unbehagen: We’re working on a technology called Shortest Path Bridging. We’ve imbued our latest generation of switches and routers with an enhanced and extended version of SPB, which we call Fabric Connect.
Whatever you call it, it is an evolutionary leap forward because it lets you completely virtualize both the networking and forwarding in your data center or private cloud. This makes the physical layout of your network irrelevant, as you can build any logical network on top. It does to the network what virtual machines did to the server.
There is far less operational overhead in a Fabric Connect environment than a traditional network, but an order of magnitude more functionality. When you have Fabric Connect-enabled Ethernet everywhere, you can do some pretty interesting things: high-performance cut-through and store-and-forward models of traffic switching, virtualization of network connectivity, migration of Virtual Machines anytime, anywhere, and more.
Moreover, we are making Fabric Connect work with an Infrastructure-as-a-Service (IaaS) technology called OpenStack that is solving real-world problems for data centers today.
What’s OpenStack and how specifically are we integrated with them today?
Unbehagen: OpenStack was created by NASA and Rackspace to solve arguably the number one problem in data centers today: Simplifying the creation and movement of applications and virtual machines in a public or private cloud. Traditional methods of configuring a network, storage and virtualized servers could take months and involve several complicated, independent steps.
OpenStack integrates several layers of modules that manage each step individually, but combines them all into a single automated interface. An administrator wants to say, “I need 15 machines with this much CPU and RAM, and this networking setup.” And that’s it. No network, compute or storage admins have to get involved because OpenStack creates the VMs and all its necessary resources for them in the background.
Networking remains the sole pain point for OpenStack users. For example, OpenStack only lets you move virtual machines in the same subnet, IP attachment and VLANs today, and the number of Virtual LANs is limited to 4,096.
We’re integrating Fabric Connect with OpenStack’s ‘Quantum’ project. Now, network administrators can have complete flexibility to move their resources around the network. That is true network virtualization, with flexibility and scale while maintaining the simplicity of Ethernet.
You said OpenStack is a real-world enterprise issue today.
Unbehagen: There are many medium to large enterprises that have been deploying OpenStack all over the world for the past several years. I recently met a number of customers using OpenStack who are deploying Avaya Fabric Connect because of the simplicity it provides. They were keenly interested in how we are deploying SPB into OpenStack because, they said, they really love both. These are government agencies, financial services, data center hosting companies, and healthcare institutions.
At Interop this week, we plan to demonstrate that our Fabric Connect technology interoperates with SPB-based products from Alcatel-Lucent, HP and Spirent. And I know of a few other vendors planning to release SPB-based products later this year.
Besides SPB and OpenStack, where else is SDN headed?
Unbehagen: There are many, many efforts under the umbrella of SDN that are very interesting. OpenDaylight, for one, has a lot of promise. In the IETF, there is something called Network Virtualization Overlays (nvo3). This is about making it easier for virtualized applications to move around the IP network. There’s also Interface to the Routing System (i2rs), which is intended to boost intelligent communication with existing routing protocols to ensure better scale and better stability. SDN can still go in many different directions. Nothing’s set in stone.
I do think SDN is a bit of misnomer. Every piece of hardware, every switch and every router, already runs on software. They are already software-defined. Ultimately, I think SDN needs to evolve and become aware of the applications on the network so that you’re managing and improving their usability. That’s why we prefer the term Application-Driven Networking (ADN).
Avaya is one of few companies that can actually pursue an ADN strategy because we deliver from one end of the stack to another. We make the software that goes on a tablet or smartphone, as well as on servers connected to storage and the network in between. That gives us a cohesive view into end-to-end application performance. And we think that’s the sort of visualization that enterprises will deem critical.