What's next with hypervisors?
- 13 December, 2011 22:28
The world of hypervisors is complicated by the fact that there are proprietary and open source tools and the latter are often pressed into service in different ways, say nothing of the fact that the whole market is evolving quickly. To get a handle on recent developments, Network World Editor in Chief John Dix corralled a panel of experts to assess where we are today and where we're going. The experts included Al Gillen, an analyst IDC who tracks virtualization developments, Kerry Kim, director of solutions marketing at SUSE, and Adam Jollans, program director of IBM's Linux and Open Virtualization Strategy.
Let's start with a stage-setting question. What are some of the key differences between the various hypervisor camps at this point?
KIM: One of the differences between proprietary products and the open source hypervisors is we're seeing a more rapid pace of innovation around open source technologies and, in conjunction with that open development model, more complementary modules and service and support options.
GILLEN: I'm the first one to agree that open source solutions tend to rev very quickly and that's important in certain market segments, but enterprise customers frankly don't want to have that kind of change pushed at them. They don't download the open source bits themselves, they go to a commercial provider like SUSE or Red Hat or Oracle and get a commercially supported product. It's not because customers don't want innovation; they just can't really embrace the innovation as quickly as it comes.
TECH ARGUMENT: Open source vs. proprietary software
JOLLANS: The big change happened around 2005 when we started to see hardware support for virtualization in x86 processor lines from both Intel and AMD. Before that, doing virtualization on x86 was actually quite difficult. With hardware support, it became a lot easier, which opened up the opportunity for many more hypervisors in the marketplace and really set the stage.
What would encourage someone to use a proprietary hypervisor versus one of the open source tools?
GILLEN: There are a couple of things that drive customers to choose a hypervisor and it's not always about the best technology. In many cases customers chose a hypervisor based on the fact that they have multiple platforms and they try to pick a common denominator; one infrastructure that will support all of the servers they wish to virtualize.
In other cases customers are going to make a decision based on the relationship between the hypervisor and the operating system they use, meaning they may chose a product from the same vendor so they have a single stack of software. In some cases you find customers making decisions influenced by their longer-term cloud infrastructure plans. So it really comes down to what their infrastructure looks like, what their background is, what their experience has been and where their roadmap is going.
Is it a given that large enterprises will end up with a bunch of hypervisors?
GILLEN: We see that happening today. In general, customers don't want to have three or four hypervisors, but we see them increasingly having more than one. Many customers that made early commitments to VMware, because it was in the market first, are adding other hypervisors around the periphery for workloads they feel are better suited for the alternate hypervisors. In other cases customers are looking at the possibility of a long-term move from one hypervisor product to another, so they begin to test and deploy another hypervisor and get some experience with it and see if it's going to work for them. As a general rule, I would argue that most customers don't want to have any more diversification than they absolutely have to.
KIM: I would echo a lot of what Al is saying, but for some customers the need for certain features on the management side will dictate the choice. Other times we see customers choosing a particular hypervisor based on their tolerance to risk and their desire to avoid lock-in with a proprietary vendor. And of course sometimes we see customers making their choice based on price or value, how much they think they can get for what they're paying. For example, some of the licensing we offer is appealing to customers because they can run an unlimited number of virtual guests with one server subscription.
GILLEN: I think that we should be clear that lock-in doesn't happen at the hypervisor layer. It may be true that the hypervisor stays as part of the lock-in, but from my point of view the lock-in is really happening at the management tools layer. If you've got a hypervisor running and you've got VMs running on top of it, in most cases you can migrate those VMs to another hypervisor pretty easily. It is the management tools you have to replace if you go from one vendor to another. Going forward I think we're going to find more and more hypervisor decisions being driven by the management tools people want to use.
JOLLANS: I think we're also going to see an evolution over time from the ability to manage multiple hypervisors to being able to manage the whole virtual machine portfolio from one pane of glass. So whether you're managing VMware or Xen or KVM, you want to be able to have a view of all the virtual machines in your enterprise.
VIRTUALIZATION WARS: VMware vs. Hyper-V vs. XenServer vs. KVM
How much of a problem will it be for a large enterprise to do business with a cloud provider that has implemented a different type of hypervisor than it uses in-house?
KIM: If the cloud provider is using a particular technology there's a natural desire to use a corresponding equivalent on your side. I don't know that that's a technical issue or more of a business perception issue, but that's certainly some of the feedback we've been getting.
JOLLANS: A lot of the time companies are looking at cloud as a flexible extension to their IT systems so they want to be able to offload workloads from internal systems, so it's going to become more and more important to agree on that.
GILLEN: Adam you're right on the money there because, over time, the attributes of the cloud that become appealing are the service levels offered, the cost per unit of measure that you're purchasing and your ability to interact with that cloud easily and seamlessly.
Let's dig a little bit deeper on the management side of things. Are there large differences in terms of the capabilities supported by the different toolsets or are they all similar at this point?
GILLEN: I think you'll find there are differences and the differences are somewhat subtle. Sometimes it comes down to, for example, do you have the ability to extend directory or federate into a cloud. Do you have the ability to move blocks of IP addresses and bring storage from one set of services to another, and how these things are implemented differ from vendor to vendor. Some management and cloud infrastructures are more suited to certain types of environments. So, for example, if you're looking at a System Center management infrastructure you're going to find it works a little better with a Microsoft environment.
At the same time, if you've got a very mixed environment where you've got two different flavors of Linux and Windows in the same infrastructure, in many cases that's the type of environment where customers have a natural tendency to look toward VMware because VMware is seen as having less of a platform agenda.
Kerry, you have a take on that?
KIM: Our business strategy has been to support customer choice in terms of the management platforms that they want to use. So we're on the operating system and infrastructure layer and support various third-party tools, whether they're open source or proprietary.
How about you, Adam?
JOLLANS: In terms of the management of hypervisors I think the homogenous versus heterogeneous is one of the key points here. My expectation is it's going to get more heterogeneous than less heterogeneous. So either the VMware tools need to be able to manage other hypervisors or you'll have to turn to tools like IBM's VMControl, which can start to manage multiple hypervisors. Now, in a cloud you quite possibly have a pure environment because you're going to optimize by standardization. If' you're an enterprise IT center it will probably be heterogeneous because the rest of the data center tends to heterogeneous anyway.
Let's take a step back and look at Xen and KVM. There had been a battle brewing between the two, but with the recent decision to add Xen to the Linux core (KVM is already there), some of that has evaporated. What do you folks make of this recent development?
GILLEN: Over the long term it's more sustainable to have the hypervisor built into the operating system simply because you don't have dual sets of development going on; you don't have to develop a set of drivers for every new piece of hardware that comes out. But commercial support for Xen has waned somewhat. Although Xen is the third most widely used hypervisor, the problem is there's no one single version of Xen used across these different places. Over time, it's going to be more and more difficult to sustain that Xen development effort. I'm not suggesting Xen folds up shop and goes home a year or two years from now. It's a safe bet Xen's going to stay around for the rest of the decade.
KIM: We've got a number of customers that are deploying production workloads in Xen, so we can't abandon them.
GILLEN: Abandonment is probably too strong of a word. I think that over time Xen becomes seen as more of a legacy solution and, although it continues to be used and supported, over time users wind up going to some other environment.
JOLLANS: I think a lot of this is about the community and about the ecosystem. If Xen diverts into multiple code bases then you don't get the same sort of community effects here. With KVM, along with SUSE and Intel and HP and Red Hat and a number of others, we put together a couple of groups recently: the Open Virtualization Alliance, which is looking at how do you educate the market about KVM, and more recently oVirt, which is about getting virtualization management to a common code base as well. So if you can hold the communities together and develop a single code base, then I think long term you've got a much better future than if it diverges into several code bases.
GILLEN: Frankly that's one of the reasons why I have concerns about Xen. Xen is widely used in the service provider community, but they're not using Oracle's or Citrix's version of Xen, they're using open source Xen. They got the bits and they modified them and they've built management tools around it. You can almost call it a fork of the Xen line because they've gone off and done their own thing and those things don't re-converge very easily. As a result, you don't have one large ecosystem, but rather a series of smaller ecosystems.
KIM: I think with the recent acceptance of Xen into the Linux kernel we're beginning to see more community support for Xen, so maybe less diverging and more converging.
JOLLANS: The other thing to remember is there is an architectural difference between Xen and KVM and I think they each have their strengths. Xen's approach is they write to their own very small footprint hypervisor, so they're writing all the core services as tightly as possible to provide that platform. The KVM approach is to leverage Linux and then strip it down from there. And depending on what you want to do, one may have a strength or the other may have a strength.
Al, who has what market share?
GILLEN: VMware is number one, Xen is second, Microsoft's Hyper-V is third.
How much of a lead does VMware have?
GILLEN: Depending on how you measure it, VMware has a little more than half of the market.
How hard will it be for the rest of the camps to make headway against VMware?
GILLEN: I don't know if that's the right way to think about it. We don't see companies going in and ripping out VMware and replacing it with KVM or something else. We see customers over time start to implement secondary hypervisors. In fact, getting VMware out once it's in is very challenging.
JOLLANS: One of the fun things about this industry is how fast the technologies move because the rapid refresh rate represents opportunities for new technology to come in. As people bring in new servers, for example, there comes the question, should we continue with the existing hypervisor structure or do we now think this is a time to change to a new hypervisor strategy? So the market shares can change and the whole dynamics of the industry move simply because the refresh rate is happening.
GILLEN: I fully agree with Adam. I think if we dial out eight or 10 years from now you're not going to see VMware has contracted to 30% of the market and everybody else is 70%. It's going to be more a matter of the hypervisor market as we think of it today will still exist, but the cloud market will have grown enough that it will become the metric we're measuring.
KIM: It's similar to what we saw with the adoption of Linux. It isn't necessarily a displacement, although we have seen some specific examples of platform migrations. It is more net new growth, organically growing the enterprise data center around the solutions that drive the business. That's why we've chosen our particular strategy to adopt support for multiple hypervisor technologies, as well as in general, pursue a strategy that incorporates support for mixed IT environments.
JOLLANS: I think an interesting question is, does the hypervisor commoditize? Looking out five, 10 years, there's all sorts of possible ways it could commoditize. You could, for example, see the hypervisor included with every operating system. There's affinity between KVM and Linux and affinity between Hyper-V and Windows, which could tend to drive that. You could see the hypervisor go down into the hardware so that every server comes with a hypervisor embedded in it, in which case it's commoditized in that aspect. Or you could see it continue to be an independent hypervisor, or a mixture of all three. These dynamics are going to shape the market and I'm not sure we see how that plays out.
GILLEN: Adam, you're absolutely right. Hypervisors will continue to commoditize and going down to the hardware is absolutely what they have to do. That's where the hypervisor technology belongs long term.
Explore that a little bit: How will that change the whole environment?
GILLEN: I would argue that it doesn't really change the game at all because it's not the hypervisor that defines the winners or losers, but rather the software that goes around the hypervisor that gives you the management capabilities. It' gives you the services and self provisioning capabilities. It gives you the capabilities of building out a single image cloud that allows your next generation applications to run. Those are the kinds of things that really are going to shape the future.
So, your cloud provider might have one kind of hypervisor, some servers you buy may come with another, and some special application somewhere else may have a third ...
GILLEN: Yeah, I think that's a fair way to look at it. What does a hypervisor do? It provides a virtual image of what looks like a piece of hardware to the operating system that runs on it, so do I really care if it's Brand A, Brand B or Brand C hypervisor. As long as it looks like an x86 server to me my operating system is happy. Where the differentiation comes in is the things that manage and do things with those guest operating systems and the hypervisor itself.
How hard is it to patch together mixed hypervisor environments today?
GILLEN: There are tools that can span multiple hypervisors, and I think it is a good thing to have a single pane of glass for all of your hypervisors. But the reality is most customers tend to use the set of management tools provided by the vendor that supplies the hypervisor, so you tend to have vertical silos of hypervisors and management tools, and if a customer has more than one hypervisor, they tend to have more than one management tool.
KIM: At SUSE we're seeing the same thing. Even though theoretically it is possible to have hypervisor interoperability, practically speaking it seems like our customers are still adopting technology in those kinds of silos.
How about you Adam, what's your take?
JOLLANS: I think it depends where they're coming from. I think if they've got a mindset of enterprise IT management, then they're looking at how to extend that downward to manage whatever virtual machines they have. Now the environment gets more complex as we get into virtualization on smartphones, virtualization on tablets and things like that. That's coming, and it will make the environment even more heterogeneous.
One last question. We've talked about the view from eight years out, but what about this coming year? What kind of developments would you folks expect to see in the next 12 months?
GILLEN: We're seeing a lot of interest around private cloud right now. Over the next year customers are going to be making some commitment to what their private cloud strategy is going to look like, which means vendors need to make sure those customers understand their roadmap for delivering private cloud functionality. So the discussion really starts to elevate above the hypervisor and it starts to be more about, "OK, tell me what you can do to fill out my private cloud story."
The second thing is that KVM has arguably hit mainstream and we see it as being very acceptable for customers to deploy. Now, will that mean a real rush to KVM? And I think the answer is no. We've been talking all along here about how it becomes a very evolutionary play rather than a revolutionary play. So we would expect to see customers who deploy commercially supported Linux deploy more and more KVM with those Linux installations. So going forward that's one of the other dimensions we're going to see happening.
The third thing that's going to happen in 2012 is that Microsoft is going to bring out the next release of Hyper-V as part of Windows Server 8. Microsoft has made a lot of improvements in their product and, frankly, Microsoft needed to make improvements to get that product up to speed. But they're going to come forward with a product that's pretty feature rich and pretty capable, and at that point I think customer that have been holding out waiting for a Microsoft based solution will arguably have a product that they can feel confident in. [Also see: "Microsoft claims Hyper-V will leapfrog VMware"]
JOLLANS: I agree with all of that. I also expect to see the virtualization management tools for KVM maturing with a number of initiatives to fill out that space. I think we're going to see hypervisors become a standard part of the operating system. And I think we're going to start to see also people exploring nested virtualization, which is something that is just come into the Linux kernel. We've been doing a number of research projects with those machines inside virtual machines which, for scenarios like moving clouds about, could actually bring some benefits.
KIM: I would echo the things that Al and Adam have said about the evolution of the tools, the coexistence of the various technologies. Without going into a lot of detail, you'll also see fruits of our technical collaboration with Microsoft that extend beyond virtualization to some of the cloud solutions Microsoft will be bringing out.
Read more about data center in Network World's Data Center section.