Tracing the direction of data centre travel
Ready to roll
All that sounds great but how many companies are ready to take advantage of that automation? Brocade views reaching the SDN ideal as an incremental process.
“If you have to buy a TV tomorrow, you don't buy one that is not HD ready. I think of it in terms of maturity,” says Starr.
“The premise behind the maturity model is that every time you buy a new device you take a step to the right – fully dynamic, billable by the second and so on. One of the places that the maturity model is taking us is that people are less concerned about brand. It's a value add, a 'solution'.
"In the network world what matters is that a network is ready for this IP backplane. From rack to rack to row to row to DC to DC, these three-tier architectures that are 20 years old don't deliver what is required.
“That's why we need things that behave like a fabric. This IP backplane is about adding automation and interaction with the surrounding infrastructure, so that you build intelligence into the network."
It doesn't end there, according to Starr. "That doesn't have to be the whole gamut of SDN. You can also build in application awareness, awareness of what's connected to it,” he says.
“There are two mentalities. There are the 800lb gorillas who say 'you will build a network this way for this application’. Then there are others on the other end that build in discovery and look for what is attached.
"Nobody really builds application-ready infrastructure anymore. Having a network that you can configure quickly is part of that. It doesn't have to be a full SDN capability; there can be baby steps towards it."
A network that detects the workloads attached to it and automatically configures itself to suit? Sounds intriguing.
Worth the price
Building in extra features that you won't necessarily use seems likely to drive up the cost. Why would a business pay for something it is not using, no matter what the potential might be?
Starr maintains that “HD ready" pieces are important for moving the network forward. He wants to see support for overlay coming to more networks.
“When people start using SDN they are going to start using tunnels. One of the key points is the visibility of flows and the ability to do something about it,” he says.
"That visibility is important. Those kind of stepping stones are important. You might not necessarily go down that programmability route, that OpenFlow route, or use those features today, but you can get some value from your infrastructure and be ready for tomorrow."
In Starr's view it is all about choice. "It is about what you choose to do in the networking world and having a commonality of management, something that can be deployed very easily. For example, our load-balancing platform will be delivered as a virtual machine or as a hardware client,” he says.
“If you get it as a hardware client you can partition it. You can say 'I will build it and they will come’. You can build a large hardware model or start with virtual machines and grow as it goes."
Starr says Brocade has very much embraced the concept of SDN, whether it is hardware from a load-balancing and routing perspective or a hardware choice.
The middle way
Some customers will need a very large platform, some will want to deploy a single virtual machine to a customer. Those things "will meet in the middle because of things like Intel's DPDK.
Even with networking gear built to work with SDN, Starr sees room for price differentiation. "I think there are differences between standards,” he says.
“You can have an IP backplane and remove a lot of the historical constraints, but where you start to see differences between vendors is depth of queue – which has quite a large difference for certain types of traffic. No drop, buffer depth, queuing capabilities have an impact on performance and cost per port."
Starr thinks SDN has a lot in store not just for the enterprise or the cloud service provider but for the mid-market as well.
If Brocade can deliver on its vision, then he is probably right. Adaptable switches that work together in a fabric but can still do it old school if needed seem like a great way to bridge the gap between the old and the new.
The data centre keeps on evolving. ®
Sponsored: IBM FlashSystem V9000 product guide