VMware CTO: Adapt, enable choice, or die
To some outsiders looking in, VMware has gone through quite a change recently.
A year ago there were criticisms that VMware was the proprietary private-cloud and virtualization leader, with a perception by some that the company was pushing vendor lock-in to ensure customers work within the VMware ecosystem. Perhaps the height of the angst of "VMware vs. the world" came when its VP of Cloud Service Mathew Lodge now almost infamously called open source cloud projects "ugly sisters" in response to reps from open source projects claiming they were more open than VMware.
GOT SKILLS? 12 effective habits of indispensable IT pros
Since then, there's been somewhat of a different tune from VMware. At VMworld this spring, the company outlined its software-defined data center concept, an idea that the entirety of data center operations, from compute to storage and networking, can all be virtualized and controlled by overlaying software.
The company went on somewhat of an acquisition spree too, to enhance what some consider a new concept for VMware: enabling multi-cloud management and migration of workloads outside of VMware environments.
Network World recently sat down with one of the men behind VMware's strategy and technology, CTO and Senior Director of R&D for VMware Dr. Stephen Herrod, at MIT's annual Emerging Technologies conference in Boston, where he discussed the future of networking being software-defined.
Note: This is a lightly edited transcript of an interview with Herrod on Oct. 25.
Network World: There's been talk by some that VMware has a new strategy around supporting multi-clouds and opening up to open source projects. Tell us in your own words what VMware's overall strategy is right now?
VMware CTO Stephen Herrod: We have to continuously be reinventing ourselves and continue to push the envelope quite hard. We've been working on this strategy of the software-defined data center for a while now, although we've just coined the phrase earlier this year.
The concept is that we've done a great job in server virtualization but whenever you go out and talk to a CIO or someone who's trying to transform everything in the data center, they'll tell you that virtualization has lowered the cost of creating a new virtual machine and we've made it easy to do this, but what about the rest of my infrastructure?
When you really get into the data center, there's a lot of rigidity. Ground zero of that is the network. It's one of the most constraining aspects of the data center today. The number of network admins that say you can only place this here, there's a firewall there -- it's all very constraining to the ultimate vision of having everything being able to move around.
It's been somewhat of a journey for us, where we've started by focusing successfully on server virtualization. Then, our partners and ourselves have been doing quite a bit around storage virtualization. I'd give ourselves and our industry a "B" at this point for making that better. But for networking, I would give it an optimistic "C." That's why software-defined networking has taken off so quickly -- it's recognizing that the same benefits of virtualization that have showed up in other places can occur in the networking space.
NW: Where are we in the software-defined networking life cycle? Are you seeing customers implement this yet, or was part of VMware's purchase of Nicira just planning for the future?
RELATED: Why VMware spent $1.2B on Nicira
Herrod: Within the VMware environment, we've been doing quite a bit for a while now. The number of virtual switches we have in our customer base makes us one of the top five vendors of switches, if you count the virtual ones. So within our own environment, we've been taking people forward quite aggressively. But there were two things that we found very interesting about Nicira.
They were created from the beginning to think about networking as a distributed systems problem, rather than a networking problem. It's very important to think about it that way if you're going to achieve the type of scale you need to be in charge of all parts of your network. This is the group that founded and started OpenFlow, so the technology side represents the best-in-class thinking of how networking is evolving. The other side is that they didn't work at all in the VMware environment and we didn't have much that worked in a non-VMware environment.
And that gets to the higher level of our strategy. Our goal is to deliver a cloud suite that makes it as easy as possible to build a software-defined data center, but we've also gotten clear feedback to enable choice. Most customers, we believe, will use the VMware environment for most of their production workloads, but they also may want to be able to interoperate with other platforms, such as OpenStack, Amazon and others. The key goal of networking, almost by definition, is that it has to connect everything. By buying Nicira, we got the technology as well as the best solution that is in already production, as Nicira has announced eBay, AT&T, Rackspace and others as some of its customers.
NW: So you really see SDN as a transformational shift for networking?
Herrod: It's pretty clear. After watching the server virtualization world for 15 years, it looks exactly like that. I think it's as easy of a bet as you can make that the network will be changed to be virtualized. It'll happen both within the data center and I think even more interesting to watch will be when it can expands across multiple data centers.
NW: How does Nicira's approach to SDN differ from others in the market, say for example Cisco?
Herrod: Nicira was really the first and is now the largest of the SDN strategies being used so far. We actually collaborate quite a bit with Cisco. The good news is that everyone recognizes the same problem, which is the rigidity of the network is keeping you from moving things around as much as you'd like. There are a number of different solutions to do that and Nicira represents a mostly pure software approach that still uses hardware on the back end. Cisco, by its heritage, is taking a more ASIC-oriented [application-specific integrated circuit] approach and using more hardware aspects.
I think the answer, where this will all end up, is similar to what we saw in server virtualization. I think you'll have a software layer that orchestrates and moves everything, but you'll have a variety of different pieces of hardware that you can plug into it that will add value and functionality on top of it. I don't think there will be a black-and-white, winner-and-loser type race. I think there will be a lot of choices customers will make that will need to fit into some common framework.
NW: So there's room in the market for strategies from multiple vendors?
Herrod: I frankly think there will be one software strategy, or at least one primary one, and we certainly believe we have that in terms of how you manage and orchestrate all of it. But I think there will be a lot of differentiation in how you satisfy that goal, and how you move the bits and value-add around security, load balancing, and all these others areas.
NW: How do you plan to integrate Nicira's technology into VMware products?
Herrod: We actually have a bunch of very large engineering meetings going on to get the precise things we'll be rolling out soon. But the strategy is that we obviously want them to be able to interoperate perfectly. Ultimately we want to be able to do two things. One is to have a management layer with very good scale that allows you to create these overlay networks and allow them to fly across VMware and non-VMware environments. So it certainly embraces heterogeneity. But where I think things really are headed is a large set of higher-level networking services, the Layer 4-7 areas such as security, load balancing and everything else. That's the next horizon that really needs to fit into this new world.
Today you statically place physical appliances on the network and things have to be connected to them. Those are going to get absorbed directly into the software-defined networks. There will be these higher-level network services that also have to follow workloads around, which means they have to be very nimble in how they move and work. So you'll see us do a lot of work taking what we've done with vShield and bringing those same technologies to the non-VMware world and fitting them in with the network.
NW: What's the time frame for that? And would you roll out the management layer before releasing the higher-level networking features?
Herrod: We're doing them in parallel. They're related but they can be handled in a distinct way. We don't have a time frame yet of when it will be released.
NW: I wanted to move on to the DynamicOps acquisition recently. It seems that VMware has really been embracing a multi-cloud support strategy, especially with this purchase. Did you feel that was necessary in response to criticisms by some around VMware pushing vendor lock-in?
Herrod: What it really came down to was customers wanting choice in where they do different things. By having choice and being able to give options to be able to better move things around, honestly, they'll be more likely to bet big on VMware if they know they can move away if we're not adding the value they're after.
At the top level, DynamicOps is a tool that helps people automatically place their applications where they want, based on constraints they may have about data privacy, or for example, requiring certain engineers to work in the least expensive infrastructure environment. Most people use it with VMware, but it also works with Amazon and it allows you to deploy to physical hardware, if you ever want to do that (laugh).
It really gives customers this great combination where networking connects all these different worlds and there is a top-level decision engine that can place things in these different worlds. I think this approach will give customers the choice to move things in ways that best suit their business needs.
NW: VMware recently announced support for OpenStack in vSphere. How do you hope to continue to support that into the future?
Herrod: We've committed resources to make that integration work quite well and you'll see more announcements as this keeps going, especially leading up to the next Grizzly release of the OpenStack code.
There's a lot of demand from customers who use vSphere today to have some choice in cloud management systems around it. We want vCloud Director to be the top choice, but if they want to use something else, we want to support that, too. It's one thing to make it work as a basic place to run and launch VMs, but vSphere has so much value around fault tolerance, high availability, SLAs and I/O controls; we want to make sure people can leverage those other environments as well. So it's more than just letting you run a VM, but it would also be even cooler if you could take advantage of these other features.
NW: Some consider OpenStack a competitor to vSphere on the private side and vCloud Director on the public side. So how do you balance your work in the OpenStack community?
Herrod: I think there's been a lot of discussion around that but it's pretty simple from our point of view: If you want to use OpenStack, we want vSphere to be the best possible way to do it. We absolutely compete in some aspects. vCloud Director competes on the public and private cloud side, but we have competition and cooperation across the entire stack that we work in, so I don't see any ambiguity in how we can enable both of those. We integrate our tools with BMC, CA and others. It's a messy heterogeneous world, and it's really hard to make that work well, but our heritage is getting into the middle of that mess and making it simpler.
NW: As a final question, here at EmTech, MIT's Emerging Technology conference, there's a lot of talk about new technologies that could be game-changers, especially for legacy tech vendors. How do you see the state of the startup community and emerging technologies that are evolving now?
Herrod: At VMworld in San Francisco we had more than 250 companies on the floor, with about half of them startups and the other half well-established companies. What's really cool is that the startups are creating their entire products and companies off of VMware deployments. That's the kind of thing that we completely embrace, whether it's in storage or networking. But I think our biggest worry is the guy in the garage doing something disruptive that we haven't thought of yet, rather than the market that's out there. So we continually do internal development and startups within VMware. We have to sort of always be paranoid, which we are, to make sure we're not complacent at all.
Read more about data center in Network World's Data Center section.