From the whatever-people-will-pay department
My colleague at Data Mobility Group, Walter Purvis, sent me a link to a Barrons blog which summarizes the thinking of several analysts on the issue of VMware’s worth.
Let a thousand flowers bloom
There is some disagreement.
What the analysts are trying to do, at least some of them, is figure out the current value of the future cash flows of VMware. Others are less analytical and just say, essentially, recent tech company sales get x times their revenue, so VMware is worth anywhere from $8 to $12 billion given its current sales.
While a sale or IPO is a point-in-time “liquidity” event, analysts who forecast stock prices look to future cash flows. Since EMC professes no interest in spinning VMware out, one has to look at the impact on EMC’s cash flows. Wall Street is willing to pay for future growth. That’s why some of the analysts are trying peel back the onion and determine the longevity of VMware’s rapid growth.
Market saturation
The analysts who appear to have thought about it agree that VMware will saturate its primary market of server virtualization in the next few years. If Longhorn works – hey, it could happen – maybe even sooner.
This is where it gets weird
They also agree that desktop virtualization is the real opportunity for VMware. Here’s one theory from a comment on the Seeking Alpha blog:
Today of course there are billions of dollars spent to secure personal computers from viruses, spyware, malware and phishing . . . . Virtual machines offer a very powerful solution to this problem by isolating different activities (like say browsing the web, or playing that multi-user video game, or even leaving the corprate firewall) in a separate virtual machine that is destroyed at the end of the users session. In effect, who cares if you pick up viruses and spyware when at the end of your session you just destroy the virtual machine taking the viruses and spyware with it. It is a super powerful and elegant solution to one of the most persistent problems plaguing PCs today. In this sense you could imagine why almost 100% of all PCs in the future could come with virtual machine technology.
The phishing reference struck me as odd: exactly how does virtualization keep me from logging into a fake bank site and giving up my password? That was problem one.
Yet I’m thinking, OK, I have a virtual machine. I download a pdf or something that I want to keep. Presumably that pdf goes on a real disk. So even after I zap the session’s virtual machine, I still have stuff on my disk that might come back to bite me. Or is that just paranoia? Somebody who knows more than me please comment.
Using virtualization to create time-sharing computer systems
Another analyst sees the desktop market differently:
. . . take away your desktop PC, and replace it with a thin client tied back to a server running many instances of corporate desktops at the same time. He thinks the average compression ratio for desktop PCs would be about 45-to-1: one server taking the place of 45 PCs. As he notes, nothing changes for the software vendors in that scenario – you’d still need the same number of software licenses for anyone using Microsoft Office – but you could save considerably on hardware costs, and dramatically improve the ease of desktop administration.
Again, I can’t judge the technical aspect of this approach, but it seems whacked on its face.
For example, according to this pdf, physical memory is the most-used resource in VMware ESX. So to get that 45:1 ration, you’d need a pretty big server, one that could support at least 32-64 GB RAM and sufficient disk for everyone.
What kind of application would meet these requirements? Call centers, sure. Office PCs that are lightly used, sure. But the secular trend is towards notebook computers and away from desktops. So this is a massive market? For how long?
As I’ve noted before, I believe VMware solves the wrong problem: putting many virtual servers on a single physical server. The long term issue is getting many servers to act as a single virtual machine.
That said, it clearly has continuing value. That isn’t the issue. Growth – how much and how long – is the issue.
The elephant in the room
And how does Microsoft play here? Are they just going to stand back and give up $12B of market cap?
I don’t think so.
Whatever the technical merits of virtualization for solving the PC world’s problems, the market reality seems stark: Microsoft gives away basic virtualization in Windows Server – and charges for specialized add-ons while selling even more Office/Vista virtual machine licenses.
Microsoft bought a virtualization company a few years ago. They have in-house expertise – and I know they have Ph.Ds – and they’ve made noises about the virtualization market.
IMHO, Microsoft could rip the heart out of VMware’s business in 18 months from a standing start. I don’t even see an antitrust downside. And since they own both sides of the server-client interface, they’d have a level of control that VMware can only dream of.
The StorageMojo take
Ultimately VMware is worth what investors are willing to pay for it. The idea that VMware is simply going to keep growing at its current torrid pace for the next several years seems naive to me. The corpses of Microsoft roadkill litter the history of PCs.
If I were tempted to join VMware for the sake of those stock options, I’d want to make sure they vested in three years, not five, and that vesting started at six months, not twelve. When Microsoft crashes the party the fun is over.
Update: I added a few lines to clarify some points.
Comments welcome. I don’t know anything about VMware, so I’d love to hear what folks think of the proposed scenarios.
Robin,
Virtual serving technology has a long future ahead, despite what the ‘guys smarter than me’ think. Consolidation is just the first application. After that look for a number of great systems management functions that allow real DR testing, what-if analysis, system roll back functionality and historical system versions, system simulation and modeling. All kinds of stuff we never figured we’d be able to do with PC systems. If you could have ready access clones with real data to work withand experiment with, how much better could you prepare for planned changes and unexpected disasters? Way! desktop virtualization mught be big, but that doesn’t mean server virtualization hits the vanishing point in a few years – it keeps going and going.
Marc,
I agree that VMware has continuing application for the very applications you list. But Wall Street doesn’t give a whoop about that.
They want growth. That’s the question I was trying to look at. Maybe I should update a little.
Robin
And I have to give here shameless plug for my employer — there isn’t only Microsoft but Linux virtualization is come to age lately. We may not be there yet in terms of user-friendliness and stability, but I don’t see any reason why full-virt on Xen (or whatever virtualization technology de jour) shouldn’t work for the above cases as well as VMware.
Yes VMWare is fine. But… Microsoft has its own virtualisation solution. Linux has an embedded virtualisation solution (KVM). There are many other free virtualisations projects like VirtualBox, Xen… Where does a company like VMWare stand in a completely commoditized away market? Together in hell with Netcape, probably…
Emmanuel, virtualisation is not a commodity. A hypervisor is an environment that has to be designed around, and VMWare has the best community of ISVs and application vendors who have spent time and money getting their products to work in a VMWare environment. Any competitor is going to have to try to convince this group of companies to spend the same resources again to get tested with any new platform.
Robin, I have a few responses to your post:
“The analysts who appear to have thought about it agree that VMware will saturate its primary market of server virtualization in the next few years. If Longhorn works – hey, it could happen – maybe even sooner.”
I believe VMWare has several years of strong growth left before they’re as ubiquitous as windows or linux. They have planted the virtualization seed is a large number of companies, and very few of them are using it as extensively as they can. Server consolidation penetration is not nearly as good as their testing environment product, and server consolidation is the main market opportunity for them.
“The long term issue is getting many servers to act as a single virtual machine.”
Some of their features carry that benefit already- you can move application servers between any node of a VMWare cluster with no downtime already. The only limitation is that you can not have one VM working on multiple nodes. This may change with the advent of high bandwidth, low latency network protocols like 10gb ethernet and infiniband, but the consolidation push these days is there not to address the problem that people can’t buy big enough servers, but the problem that the wintel boxes they own are not being properly utilized.
“IMHO, Microsoft could rip the heart out of VMware’s business in 18 months from a standing start.[…] since they own both sides of the server-client interface, they’d have a level of control that VMware can only dream of.”
See my answer to Emmanuel. 18 months is a very short time period to get all this software tested and configured to work with a new virtualization platform:> http://www.vmware.com/vmtn/appliances/directory/
Robin,
Your assessment of memory in the virtual desktop example is slightly off. ESX server includes a technology called Transparent page sharing(TPS). In a nutshell ESX eliminates the use of redundant copies of memory pages. This allows desktops and servers to utilize less memory on a virtual machines then on a physical machine. It is particularly useful when you have multiple machines on the same ESX with similar workloads. Combine TPS with the ability to over commit memory, and you have significant savings in memory. In my organization we currently run approximately 30-1 on our esx hosts (general windows servers). A typical host has 2- dual core processors and 16GB of RAM.
Regarding TPS in VMware, it’s still a weak point; Virtuozzo does substantially better than VMware with density of virtuals. VMware is in a really tight space, getting tighter all the time: squeezed by Microsoft on one side, Xen on another, and then technologies like Virtuozzo/OpenVZ and Solaris Containers/Zones on yet another side. Where I work is heavily deployed with VMware right now, but I expect it’ll only be a legacy in three to five years. Even today, the only new VMware deployments we’re going ahead with are when we need to put multiple operating systems (such as Linux/Windows) on one machine.
I am also skeptical about some of these desktop virtualization scenarios.
The idea of using multiple Windows VMs to improve security sounds nice to geeks, but I suspect the usability will be so bad that regular users won’t be able to handle it. Keeping track of what goes in what VM, or even what VM you’re looking at right now, will be a pain.
I think the thin-client+virtualization idea (called VDI by virtual insiders) is tough to swallow for cost reasons. Thin clients cost almost as much as PCs, server hardware costs much more than desktop hardware, and then you have the virtualization software itself. If your server costs 30X as much as a desktop PC and you get 30:1 consolidation, you haven’t gained much.
Disclaimer, I am fairly big fan of VMware and have been using it for a few years. While I tend to gravitate towards MS technologies, I think that you need to pick the appropriate technology for the problem at hand.
I think that VMware has strong growth potential over the next 5 years, beyond that is anybodies guess. A recent Gartner report predicted the number of VM’s to rise from about 500,000 currently, to 3 million by 2009. They have already established themselves as a market leader with the widely successful VMware appliances, and establishing alliances with Dell. A few responses to your post:
“For example, according to this pdf, physical memory is the most-used resource in VMware ESX. So to get that 45:1 ration, you’d need a pretty big server, one that could support at least 32-64 GB RAM and sufficient disk for everyone.â€
In my opinion VMware isn’t really a played in this game, although they would like to think they are. If you are going to run 45 instances of XP on VMware ESX, there are better solutions (MS Terminal Services and Virtuoso to name a few). The few examples your cited are small market segment, so they don’t really have too much impact on potential earnings or growth for VMware.
“And how does Microsoft play here? Are they just going to stand back and give up $12B of market cap? I don’t think soâ€
They really don’t have much of a choice. There technology is 5 years behind already. Virtual PC 2008 (MS answer to VMware) will not support hot migration (aka vmotion). In my opinion, this is a requirement for any virtualization technology, and has already saved us when we experienced some bad hardware. This product, which will ship sometime in the next 6 months, won’t even put MS on the same field as VMware. The only thing MS has going for it in the virtualization market is the licensing.
Regarding Emmanuel’s comments, I completely agree with Basil. It is much more complicated then that. The core virtualization technology may be a commodity, but the features and abilites of VMware are what make it stand out.
We have only a small VMware (7 ESX servers) infrastructure in place supporting about 40 VM’s, with lots of room for growth. However, it has already saved us probably 10’s of thousands of dollars and hundreds of man hours.
Andy,
What you see as VMwares weakness, in my opinion is its greatest strength. With VMware you eliminate the need to know 3 different systems, and all the ins and outs that go with each system. This considerably lowers training costs, learning curves, and the potential for human error, which is by far the greatest cause of downtime. While I agree the other products are definitely superior in terms of memory utilization, they are not nearly as flexible. With these other technologies, you need more Ethernet ports, SAN ports, HBA’s and LUN’s. Remember, ever year memory gets cheaper and servers add more slots to fill them with. Two years ago, 2GB DIMMS costs thousands of dollars, now you ca get 2 for $795 from Crucial.
Brad,
Not sure what you mean by “With VMware you eliminate the need to know 3 different systems” and “With these other technologies, you need more Ethernet ports, SAN ports, HBA’s and LUN’s” – can you elaborate?
I may be misunderstanding what you meant (apologies, if so), but we still need to know the OS of each deployed OS – and we’d be deploying those OSes whether or not we were using virtualization. In fact, what I see as one of VMware’s greatest weaknesses relative to Virtuozzo or Solaris Containers is that it doesn’t really give you the tools to simplify OS management of your guest machines, and despite the good efforts VMware has made, OS deployment is still much faster with Virtuozzo and Containers.
As far as ports and LUNs, we actually see greater reduction with Virtuozzo in particular: We get better density with it – my favorite example is the eleven production virtuals we have running on a dual-P3 with 2GB of RAM. Further, Virtuozzo’s shared-storage-free live migration on Linux lets you do away with the SAN entirely, if you so desire.
VDI has to be done in the right way. The thin client really has to be a thin client (not a “desktop-on-a-desktop”).
Check this out:
http://uadmin.blogspot.com/2007/01/sun-ray-x4600-vmware-esx-stories-from.html
Any,
By 3 different systems I mean 3 different ways of managing virtual machines. To manage Solaris VM’s via containers, you obviously need to be familiar with Solaris containers. Since Solaris containers can’t be used to for Windows or Linux VM’s you need another solution such as Virtual PC, Virtuozzo, or Xen to name a few. Adamantly, if you are a competent Solaris admin, containers shouldn’t be too much of a stretch for your administrative skills, and the same goes for Virtual PC. In this scenario, you would need 1 machine for Solaris (Containers), 1 machine for Windows (Virtual PC) and 1 machine for Linux (Xen). This could be reduced to 2 systems (i.e. software packages) using Virtuozzo to manage both Windows and Linux.
If they do offer HA similar to VMotion, then you would need to double the physical boxes to 6. I am not familiar with the HA offerings of Solaris containers or Xen, so I don’t know if having a second physical box would make any difference. If Solaris on x86 can be used then 2 VMware ESX servers could accomplish all this.
In terms of the increased number of LUN’s I was referring to the fact that each of the solutions provided above would require its own LUN(s) formatted with its particular file system. Again, with the solution above, it would require more Ethernet ports, HBA’s and SAN ports then 2 VMware ESX servers. Regarding the SAN free live migration from Virtuozzo, this is a nice feature, but obviously doesn’t scale. If you have fileservers as VM’s (which we do), it isn’t realistic to migrate hundreds of gigabytes back and forth when you need to take downtime on a physical host. Correct me if I am wrong, but Virtuozzo doesn’t offer a live migration strategy for Windows, do they?
Hi Brad
Solaris Containers are lightweight virtualisation, which means they share the same kernel. This means, that there is no such thing as VMotion…(would be very hard to achieve, but every virtualisation technique comes with a price…).
Anyway, the Sun Cluster software now supports the failovering of a Container to another host. The container just has to be on shared storage.
Your fear is a natural one for someone who doesn’t know anything about the actual use of VMware and the current competitive playing field. I thought the same way until I started asking questions. You do not understand how much of a lead Microsoft has conceeded VMware already, how dysfunctional Microsoft is, or how much regulators have interefered with its business. Everyone loves to point at competing virtualization technologies. Call up 10 CIOs or datacenter admins you know and ask them this: You have to purchase and deploy – in a production datacenter, at scale – virtualization technology in the next 12 months – what are your choices? Your job depends on it! There is only one answer – VMware. They own the market, having a 9 year headstart, and Microsoft is nowhere effectively until 2009. To qualify for real use, people want to play with this stuff for a year or more. VMware’s the only thing qualified in any significant way right now. It’s going to be all over by the end of this year. 85% of VMware license sales are now indirect – this a fully developed channel. VMware is being adopted en-masse right now. In the next 3-6 months you will see offerings from middleware vendors that allow you to deploy server hardware with VMware, middleware and NO OPERATING SYSTEM (well, really the OS is VMware). They’re in beta already at live customer sites. No one is going to pay for something they don’t need. Right now, they need VMware. It changes the economics of datacenters – lowering the cost of offering hosting services by 25-30%. CIOs can choose between spending $10mm to expand a datacenter at capacity, or $500,000 to consolidate existing servers with VMware. They’ll pay all day for VMware. It should be standard on x86 servers already. There’s no reason any business should purchase one without it. With enough penetration, the cost for people to start futzing around with Microsoft’s product, or anyone elses, instead of just sticking with what’s working, is not going to be worth the effort. Yes, it’s true that Microsoft could have and should have offered this technology. But they didn’t. People need it enough to pay more than $5000 per enteprise class license – which they are doing right now. VMware enhances productivity of the technology investment – pick all the nits you want over whether it’s good for AV protection or whatever. The objective fact is that anyone who does hosted services will tell you VMware cuts their costs 25-30% and businesses are paying $5000+ per license for VMware. They’re getting value in excess of the cost.
All,
Great discussion, both technical and business. I thank everyone who’s taken the time to contribute.
With such knowledgeable people commenting I hesitate to wade back in. But what the hey. There are worse things than looking like an idiot. Right?
Marc, if Google can be trusted, is an equities analyst and in a past life a serious bit head. All around smart. Like all StorageMojo readers.
I’ve been meaning to post about the issue of data centers. I came across a Greg Papadopoulos preso that noted, among other things, that enterprise data center demand for computes is now less than what Moore’s Law provides. Which is an interesting take on the virtualization market if Greg is correct.
Thus another way to look at VMware’s market is whether the enterprise compute demand will continue lag Moore’s law, stoking the demand for virtualization.
I have no doubt that CIOs hope it does. The LOBs may have different ideas, but the force is with the Glass House in the short term. Looking back at the assault of PC networks on the GH, even if the LOBs discover a massive cycle-sucking business app, it will take at least five years for it to penetrate the CIO’s inner sanctum.
So maybe VMware does have have a good five year run ahead of them.
Robin
Your comment on enterprise datacenter demand for compute is worthy of more digging! It does not agree with demand for overall datacenter capacity, which appears to be quite tight. But those two things aren’t necessarily mutually exclusive. Do the studies showing demand for CPU below Moore’s Law account for the underutilization of raw CPU cycles by the software running on them?
Another way to think about this is- If I have a datacenter which has reached maximum core density (can’t fit another blade or server without a meltdown or smushing them together), what’s the utilization rate of total CPU cycles inherent in my software and does my reluctance to build another identical data center coupled with sub 100% CPU utilization get interpreted as a lack of demand for CPU? Assume that the cost of rolling out VMware
BladeSymphony 2000 also supports Hitachi’s embedded virtualization technology, HVM, resulting in a blade server you can deploy with confidence for your most mission-critical applications.
The virtualization capabilities are on an ASIC; not software. Everyone go to the Hitachi web site and look at this technology. I believe that the ASIC way for virtualization is the way to go.