Martin Casado, VMware's chief architect for networking, spoke with Network Computing recently about the launch of VMware NSX, the network virtualization platform that grew out of Casado's startup, Nicira Networks.
The discussion included the origins of Nicira, potential issues with the overlay model, Casado's ambition to change how networking is done, and why energy and chaos are good signs.
What led you to start Nicira?
Casado: I used to work for intelligence agencies in 2003, working on pretty secure networks. Market forces don't produce technologies with correct security posture. The government took computers and programmed them to address their threat environment, but the networking gear was exactly as you got in the box. There was no programming model, and the functionality was in hardware. I thought, 'Wouldn't it be great if you could get programmability in the network?' I did my Ph.D. at Stanford, and the thesis was how to make networks have the same properties of compute.
VMware NSX uses an overlay model to virtualize the network, but there are competing SDN architectures. How do you see this battle playing out?
I think we had architectural debates because no one had anything. It's like companies having two cars and arguing about the engines rather than just racing them on a track. Now we can start actual testing. We've moved outside of fighting with slide decks.
Are there any prerequisites for a physical underlay network to provide an optimal environment for an overlay?
This will work on any type of hardware and any network. You can build the physical network however you like. The primary requirements are throughput, low jitter and other traditional physical characteristics.
More bandwidth is generally good. One reason networks are oversubscribed is that to manage VLANs and ACLs, there's a lot of configuration state on the network. If you use a lot of multipathing, you limit redundancy to one or two paths. You aren't taking advantage of as much bandwidth as you could.
The more wires and paths you have, the more available bandwidth you have. But the more roads you have, the more configuration state you have to put on these paths, and that becomes more onerous. If you put configuration state at the edge, you can build networks as rich as you want.
What do you say to the criticism against overlays not having enough visibility into the underlying network?
I contend it's much easier to debug and troubleshoot an overlay network. Today we try to put configuration state in the network. We have static routes, VLANs, ACLs, and so on that complicate it. In an overlay network, it's easier to look into these. We can pull in the entire physical underlay view. Here's a path in the physical network, what virtual paths does it affect?
We use standard interfaces, so traditional tools can read physical and virtual networks. And because we're building a software system, we expose APIs that can be used by tools. We can provide very sophisticated views for physical and virtual and map between them.
Given the vast sum of money that VMware paid for Nicira, and that you're now part of a very large corporation that has sales targets and shareholders and Wall Street to satisfy, are you feeling any pressure as NSX becomes generally available?
I am super excited. I don't know if that means pressure. This is the opportunity of a lifetime. When I started Nicira, I didn't know where it would end up. If it wouldn't be a standalone company, it would be a software company, so VMware was a natural place for us.
Has there been a cultural adjustment for you to go from a startup to a large corporation?
It's a different environment, but I'm really enjoying it. My goal is to change networking. That's far more than creating technology. Creating technology isn't the hard part. You have the idea, you have to do technical engineering, but then you have to get it in people's hands.
You need to understand branding, you need to build a support team, a sales force. These things I wasn't exposed to in a startup. I'm absolutely fascinated and interested and engaged in how you bring disruptive technology to market. VMware is different from Nicira, but it's exactly where we need to be to change the world.
It's not unusual for a startup's founder to move on after an acquisition. Do you see yourself doing something new in the near future?
I'm here until the job's done.
What indicates to you that the job is done?
Let's say broad adoption and acceptance [of NSX]. We're still in the early adoption curve. And so this should become table stakes. It should be the obvious way of doing networking.
Where else do you see interesting work going on?
Outside of the work we're doing here, I love this trend we see in large data centers where the physical network is becoming so simple. Functionality that was network functionality is being implemented on x86 at the edge. It appeals to my aesthetic as a systems designer to have features and functions built in software, and have the hardware do what it's supposed to do.
Second, whenever new technologies get adopted, people don't know what to make of them. When compute virtualization was discovered, the value proposition was consolidation. It was only later you could do DR, HA, virtual machine motion.
I think we're hitting the inflection point for network virtualization. The first value proposition is speed of operation. Now we're seeing use cases like people do full modeling, dynamic runtime optimizations, things I didn't conceive of, so it's exciting to see growing sophistication in its use.
What do you make of the open source efforts around SDN?
Having been involved in the SDN community since the beginning, our team was the first to open an SDN controller. Now you have Floodlight, Beacon, OpenDaylight and others. I think open source development and community is just a good thing. All the energy and chaos is great. Because we're seeing an industry in transformation, the more chaos and development we see, the more healthy it is.
We are OpenDaylight members, but it's too early say anything about results. But it's exciting to see all this energy being poured into SDN.
Among a segment of Network Computing's readers, you're regarded as something of a technology luminary. Do you see yourself that way?
(Laughs) I think there's a broader movement going on. People feel it in the air, they feel tectonic shifts. I've been fortunate to identify some of the major cracks, but I feel like so many of us are on this vessel together and the dialogue is great. I don't feel like a luminary, but I feel like I've been lucky enough to play an important role in articulating this change.Drew is formerly editor of Network Computing and currently director of content and community for Interop. View Full Bio