At the third and by far the biggest VMware's annual VMworld convention last week, we grabbed the chance to speak to the company's virtualisation visionary and co-founder, Mendel Rosenblum. Where does he see the company taking this fast-evolving technology?
Q: What combination of factors has made virtualisation the hot topic of today? A: A lot of the excitement around virtualisation stems from problems in the current software environment - a combination of modern operating systems and applications. People weren't happy with issues such as reliability and security. While the OS is supposed to be in charge of the hardware, people spend too much time managing them, and they're not robust enough to run multiple applications. What was needed was a thinner layer to do the mapping onto the hardware resources.
For example, you could imagine a distributed OS that allows free flow of information between applications, or one that stops bad processes bringing down the entire OS -- but we evolved not to do that.
And some of the older technology was too slow -- now the hardware has arrived.
Q: It sounds like 'modern operating systems' is VMware-speak for Windows. A: Windows is guilty but Linux has its limitations.
Q: What's the most promising development outside of VMware that you believe can or will aid VMware's stated aim of virtualising everything? A: Finally we're getting hardware support for virtualisation -- we can only do so much in software to get resources treated as a pool -- with boxes connected by faster networks that's exciting. We've never done anything in hardware, only software. Virtualisation is an incredibly useful technology -- virtualisation will be pushed by partners from shops that service SMBs and want to be able to build a virtual infrastructure. and there are vendors who want to use the technology to improve security.
Q: To what extent is VMware the beta-tester for functionality that eventually ends up in hardware -- especially the CPU? A: The role of hardware is to do stuff fast -- for example, memory page mapping. The primitives are getting faster and we have some smart people who know how to do it in software. Some of it's quite simple but complex stuff is not a good idea to put in hardware -- there are million of lines of code in VMware. But it's a pretty small part of the virtualisation infrastructure these days.
Q: What are the key criteria for deciding which virtualisation features should be in hardware or software? A: Intel VT is an example of how to do stuff in hardware -- safe, efficient without intervention of the virtual machine monitor. We didn't have that in the original software - could take advantage of having a VMM. Pass-through I/O can work but you still need a virtualisation layer that you can do good things with -- I spend my time trying to educate people on that.
Virtualisation is for running multiple things on one box - and you need a flexible system for that. Static partitioning -- people have done that -- is putting more virtualisation features in hardware but it's not the kind of virtual infrastructure that allows you to use it in a flexible way, such as Vmotion.
Q: What are the barriers to running 3D graphics in a VM - and when do you anticipate overcoming these? A: We can do some of this in software - need to wait until the ATIs and nVidias see virtualisation as important. I know they're working on it, and today's GPUs are like mini-supercomputers. The technology is not that different from Silicon Graphics old systems with multiple graphics cards -- it's just about putting the pieces together.
The good news is that even Microsoft is helping by the way it's specifying the graphics level -- you can read the state out of the chip. As for timing, that depends on graphics people. It's an example of what we'd like to see -- let the hardware architects figure out how to make it happen.
Q: We talk about virtualisation everywhere -- how realistic is that? Does it mean the death of the OS? A: No, it doesn't imply the death of the OS but it's a real opportunity for someone to do OSes for appliances.
Q: Your model seems to be to keep ahead of Microsoft by scattering a pile of free software in your wake as you push the technology forward. A: Want to avoid being in a niche - we're better than the competition, who can't do what we do so we can charge a premium price. Guided by the notion of the virtual appliance model, we needed a free player to seed the VA market. Our sales people hate it because it's free but developers like to impact a lot of people so that's a good thing for them.
But then, they also like to have stock options that go up in value.
Q: What have you gained by being open? A: We try to be fair. Our partners can't say we lied to them, and we do what we say we're going to do. We're trying to be the good corporate citizen -- though we still fulfil our ultimate responsibility to stockholders by doing right thing by partners and customers.
Q: Are you surprised at the success of virtualisation in general -- after all it's a 40-year-old technology -- and VMware in particular? A: I go back and forth - when we started I thought it would be cool if we could make this work. I don't think that thought was misplaced. The buzz was the same with Java but I thought it was just a good programming language -- I mean, I was pleased that Sun was able to do it, but it wasn't world-changing.
I am surprised it took off at such a tremendous pace but we thought it could go high as the values climbed. We looked a customers and thought 'we can help'.
I'm more than ever convinced that luck plays a big part, but it comes down to individuals who make things happen.