Enabling the on-demand data centre with SDN
Network disruption on its way, says Brocade
Interview The hybrid cloud is becoming a notable thing. Companies such as Microsoft are pushing visions like CloudOS while work patterns change to allow more flexibility in worker location.
Traditional networks focused on situations where workers perform all duties on premises owned by the company. How will that change?
I sat down with Brocade's Julian Starr to see what the company thinks our networks will look like in the future.
My vision of the future involves flying DeLoreans, Elon Musk as Prime Minister of Mars and secure IT that doesn't need me to keep punching arcane strings of code into some primitive command interpreter just to get access to a workload on another port.
I also tend to envision IT advances delivering more bang for the buck, even to proles like myself.
I am eager for Starr to validate at least one of dreams, but he begins with a caution. "The question of how you get there is an important one. We can look at it in terms of waves of disruptions that have occurred,” he says.
“Those disruptions – server consolidation, storage virtualisation, VDI, the way people build applications – all of these have had an impact."
Starr believes we are on the cusp of another major wave of disruption in networking. He thinks people will care less about the details of the network configuration, they will just care about flows.
“The applications just interact with the network. The network advertises capabilities, the applications will ask for certain types of features,” he says.
In principle, I like the idea. I stand up a bunch of networking resources and I can either assign them in aggregate or let the apps grab from the pool.
This concept of feature requesting and advertisement is a critical part of Starr's future network. He sees applications sending the message "I'm an application not sensitive to latency, I'm a high-value transaction, I'm very sensitive to latency" and so on to the network and receiving the appropriate resources automatically.
Starr feels that advertisement and automation go beyond the traditional configuration of network ports. An advertisement by an application could cause a SDN to stand up virtual machines with firewalls, IDS and load balancers that allow that infrastructure to be deployed quickly.
These automated elements could create connections between data centres as required and set up elements for monitoring, analysing and reporting various metrics – network orchestration writ large and controlled by an application advertisement.
Talk of the town
Starr says this is already happening. "We are beginning to see customers tying things into service desks, automating things, building tick boxes where people say 'yes, I need high capacity, yes I need DR, yes I need latency sensitive’. You deploy that into the environment where SDN allows you to build this on the fly," he says.
Starr gives a good pitch about Brocade's vision here, but determining whose apple pie recipe is best is still difficult when so few of us have had even seen the pies outside of a conference.
Still, Cisco is against it unless it controls it. As with Oracle, whatever is bad for Cisco is probably good for everyone else.
Turn up the bandwidth
However much fun the higher levels may be to play with, the infrastructure of the network needs to "just work". Starr would like to see a world where you can just plug in cables that allow you to instantly add bandwidth to links and don't require you to go in and configure that.
"It's important for the orchestration model that the underlying network should be this flexible," he says.
He talks a lot about the on-demand data centre. “A lot of the work we do in terms of briefing the industry is all about how this comes together – how our portfolio enables it, uniting the virtual and physical,” he says
“Creating this SDN capability whereby you can orchestrate, adding services very rapidly, is undermined if you don't have the capability on the underlying hardware."
It is all well and good to add an orchestration command line interface on a switch, but you need to look at the network capabilities, the underlying fabric, how the flows move.
One feature at a time
A lot of this can be done with what Brocade has on the table today but some of the needed bits are still waiting to be born. Sadly, there is not a lot of "get more out of what you already have deployed". Instead there is a great deal of "buy our new shiny."
Brocade is big on selling the idea that getting to this true software defined data centre is a slow process of introducing newer networking gear to your network and turning on new capabilities one feature at a time. That is why it designs its switches to be able to run traditional workloads and SDN workloads on the same ports at the same time.
"How do people move from a world where they have a network that is old, it's rigid, it's tight, it's fixed, it's three tier, to one that is dynamic, that has a backplane?" Starr says.
"It's a step change. You can't have a three-tier backplane and then have SDN the next day. There are ways people need to get there and it is not just adding small bits of software.
This brownfield deployment option raises a lot of questions
"Our strategy is to introduce SDN-like features that get people moving towards SDN – certainly having hardware that's ready to embrace it. To use an expression, it is 'HD ready': it is there, enabled and ready to take you forward. It does have that discovery capability that works with the hypervisor."
So a transition mechanism, of a sort, exists. If you buy Brocade today you will get two options: block-replace what you have and go straight to the good stuff or do it piecemeal and slowly assemble a sexy network over time.
This brownfield deployment option raises a lot of questions to which I did not get satisfactory answers. How will this all work if you are trying to transition from a Cisco or Juniper network to Brocade?
How well will Brocade's gear interact with other SDN gear? The answers will ultimately determine the viability of this scheme.
Promoting incremental movement towards the future, generational intercompatibility, standards and the adaptability of SDN-capable switches is the philosophy that is supposedly going to drive Brocade's future endeavours. This is the value it sees itself as offering in a future where the tin the network runs on is increasingly commoditised.
Pile 'em high
How commoditised is the real question. Cumulus is a vendor making a big splash by selling a lot of the same vision but doing it on cheap Chinese tin.
It seem inevitable that switches will start to be generic kit running a hypervisor with various bits of software doing their magic on packets running on top.
According to Starr, I am not far from the truth. "We are starting to see that already,” he says.
“Eventually that is part of what's going to happen. It is part of what you see with OpenFlow, where you move intelligence out of the switches and into a manager, where you have a client on the switch that receives routing and control information from a central controller.
“We are already starting to see white-box vendors producing switches with SDN capabilities in them and small startups producing network operating systems."
Starr sees this as a route to diversity of choice. "We say 'look, customer, you can have the best of both worlds'. We produce the hardware and we do soft switches with full AVX,” he says.
“People will want to build best-of-breed infrastructures, but will also want systems with a hard switch over here and a soft router over there. It is happening already."
Again, grand vision statement stuff and very much in line with the sweeping butterflies and unicorns pushed by others. When it comes to the price question, Starr's defences are up; this is not comfortable territory.
What's the use?
To sell a vision you need possible use cases. For SDN, Starr is happy to oblige. "One of the use cases for SDN is the campus guy who has 30,000 switches to manage. Does he want to deploy changes to all those switches or does he want some sort of configuration software that manages flows and rolls out configurations based on network flows?" he says.
Thirty thousand is a lot of switches. Is this completely out of the small and medium business (SMB) ballpark? Starr doesn't think so but most SMBs seem unlikely to be good candidates as early adopters.
"I guess it's a question of scale. Some organisations will make more use of SDN than others. I see some of the earlier adopters going for cloud hosting – those organisations that are reaching the limit of networks today,” says Starr.
“VLAN implementations, overlapping addresses, those people are early adopters. We've seen some early adopters in the programmability space as well. We've seen some interesting wins and customers in that space."
Are networks totally off the books? Starr thinks not. "Brocade has built in subscription models. This makes networking a utility model,” he says.
“Pay for ports on a monthly basis with very little risk – great for private organisations that want to change how they buy. Put a router in your environment, pay for it per month."
He sees this subscription model as important from taking those "data centre features and bringing them into the wider switch set”. It is a way to get the SDN capability everywhere, even into networks that are not quite ready to turn the new features on just yet.
Reg readers may be little disconcerted at the concept of being Office 365ed for basic infrastructure.
When your clients sit on servers, switches, endpoints and even software for a decade, a subscription model for basic infrastructure seems rather the exact opposite of more bang for your buck.
Build your own
Starr talks about some interesting features that Brocade has today which he thinks are great foundations for that easy-to-use automated network of the future.
The first is Brocade's one-touch network. "We have a demo we do at trade shows called 'build your own ethernet fabric’,” Starr says.
“People just walk up, plug a bunch of switches together and they just work. It's like building a network with very little interaction. I can build a network with two commands on each switch. It's an intelligent network."
Starr also touts Brocade's on-the-table-today automation. "We can work directly with VMware and discover virtual machines. I can log into a switch and see the hypervisor, the port groups and all the virtual machines it's connected to," he says.
Organic network growth is a frequent bugbear for sysadmins, making some of the infrastructure claims particularly interesting.
“You can connect up the switches any way you like. You never create a loop. You can dynamically add bandwidth. As you add cables you add bandwidth," says Starr.
Starr also touts Brocade's VXLan capabilities, as well as the ability to go into tunnel traffic and make decisions on it.
Starr's passion about all of this is a little infectious. “The network has become exciting again. We're right at this inflection point, we're right at this big change,” he says.
“I think Brocade, through culture, through heritage, the fact that we care about what is at the end of the network (storage and so on) – we find ourselves right at the forefront. We find ourselves in an exciting place. The next three years are going to be a lot of fun."
Pay as you go
Starr is eager to convince Reg readers that the value of SDN justifies radical investment or even changes in how admins buy and cost out their infrastructure.
In particular he likes the idea of charging for usage. That sounds like it might be great, if only everyone else weren't saying the same thing.
If budding cloud service providers add up everything they are expected to simply bill to the user of the network nobody is going to be able to afford to actually do anything on that network.
The economies of scale that are supposed to be the attraction of cloud computing have become aught but a red flag for large vendors.
They discuss concepts such as chargeback and subscription models as a means to make accountants happy: you buy X amount of these things, bundle them together and sell them on to the client. You pay only for what you are selling – give or take, based on the length of your subscription contracts – and you take a rake off the top.
That is great in theory, except that there is no room in this grand plan for the cloud vendor, or the enterprise customer, to realise any margin. All of the margin is being sucked out by the vendors under the premise of making the costing model easier.
Vendors then use the price of the subscription model over an arbitrary refresh period to jack up the cost of the base infrastructure so that you can't simply go buy your kit and sit on it for 10 years like we did last time.
Before those with the money to invest are completely sold they will need to see clear pricing that makes this pre-canned unicorn meat a better bet than Cumulus or simply rolling their own switches.
Brocade's SDN vision is exciting and may legitimately be the best take on SDN this Reg hack has heard so far. If Brocade can deliver, it will be a serious competitor to the other big players in the market.
Personally, I am intrigued enough to want some of these switches in my lab so I can tear them apart to see if this is a flight of fancy or the future.
What is your take on the future of networking? There is a lot to chew on here so your thoughts in the comments, please. ®
Sponsored: Hyper-scale data management