Server Virtualization Trend: Just Starting or Almost Done?

Something has been bugging me about the market share numbers for server virtualization. Is the trend is just getting started, or is it almost finished? The numbers I’ve seen say that under 10% of all X86 servers have been virtualized – maybe 7-8%. By that measure, the trend of converting physical servers into virtual ones seems to be quite early.

Things look very different when I look at the percentage of total servers (physical plus virtual) that are virtual. Most customers seem to run at least 8-12 virtual servers per physical, and some are pushing past 30 towards 50. Let’s use 10 as a conservative number, and do the math: For every 100 physical servers, 7 are virtualized, for a total of 70 virtual servers. That makes a total of 163 total servers (70 + 93), and almost half are virtual. If we are half-way converted, then the virtualization trend must be very far along, because the second half will probably convert much faster than the first half.

It sure seems to me that looking at total servers is the right thing to do, as opposed to just counting how many physical servers are virtualized, because to a user, it shouldn’t make any difference whether their server is virtual of physical. (That’s the whole point!)

On the other hand, it doesn’t feel right that server virtualization is so far along. Most customers I talk with are just getting started. Only a few have seriously converted. My math must be busted, because there’s no way that we are half-way converted.

I think the problem is that the math assumes that there is a fixed-size pool of servers that people are converting from physical to virtual. It seems more likely that cheap and easy-to-provision virtual servers will lead to a massive increase in the total number of servers. That is always what has happened when a computing resource gets much less expensive. We didn’t just replace workstations with PCs, we gave PCs to everyone, instead of just development engineers. Likewise, the cost per gigabyte keeps dropping every year, but instead of buying fewer gigs, people keep storing more and more, and their budgets stay roughly flat.

Given the history of the computer industry, it seems unlikely that server virtualization will drive costs down in the way people think. Instead, it seems much more likely that costs will stay roughly flat, but there will be a radical proliferation in the number of virtual servers. They are just so fast, easy and cheap to deploy, it seems likely that most IT shops will hand out scads of them.

I’m not sure whether to think of this as a prediction, or a warning. I guess if you get value from all those virtual servers (just like we did from mini-computers, workstations, and then PCs), then there’s no problem. But if IT shops really want to use server virtualization to save money, then they had better be extraordinarily disciplined.

Comments

[Paul Monaghan]

Dave, I think your logic is right on the money. Many, myself included, haven't undertaken a large project to P2V existing servers, but they are deploying new within VMs. While we're looking at P2V, particularly for older systems, it is no a small undertaking. However, as you've said, net new servers are dead easy to deploy virtually.

We share the same concerns, deploying new systems virtually is almost too easy. I think the ease presents the biggest challenge to the point where abuse is the only possible destination. We're all heading to a time where we're going to wake up and realize that we now have 500 servers to manage across 50 physical hardware boxes. Rather than 100 boxes running 100 servers. Those 500 server images have the same "care and feeding" needs, hardware aside, as they ever did. It's the operating systems and applications that consume most of the operations staff time. The VMs still require patching, monitoring etc so as the VMs increase that need increases.

The same can be said of Flexvols on Netapp filers, it's just so easy to create a new file system, so why wouldn't you? It's this ease of deployment that ends up resulting in an infrastructure that will end up being more difficult to manage rather than easier to manage.

Virtualization technology, of any kind, is constantly making things easier, but at what cost? I suspect only time will tell.

Paul

[Bill Taroli]

I would agree that looking at the total number of systems provisioned, and not just the number of physical boxes, is a better way to observe the trend. I also agree that this is not a static affair. I generalize the scenario by saying simply "People are Gold Fish." (People are also Locusts and Sheep, but I digress.) We will fill any space, use any resource, even if we survived quite nicely before such resources were available. In computing, we have seen this play out in cpu speeds, memory capacities, storage capacities, and now combining each of those with the ease and power of being able to easily and quickly scale our virtualized pool of systems.

I think we are still very early in the days of virtualization. While it is quite established now, we've really only seen large usage in /adding/ VMs to our already existing physical environments. This huge influx of systems has signaled the need for serious management tools to handle the onslaught, and was triggered in part by our realization that VMs don't really come for free. Virtualized environments have to be managed just as any other precious resource in our data centers.

Still there is a trend toward the thinking "but VMs are free! I want more more MORE!" Having run an almost completely virtualized support lab environment for three years now, I have been able to implement controls that keep this type of unchecked growth under control. I have also observed others who, using tools like Lab Manager (formerly Slingshot), put their virtual destiny completely in the hands of their users and have repeatedly paid the price -- there's nothing like seeing a 25 node ESX cluster with 16TB of storage drop to it's knees because people's hunger for systems outstretches capacity.

As we get better controls in place for managing VMs and the tools catch up to the basic technology, I think we will begin to see a trend toward P2V in the data center. Already there are some who have done this aggressively, even for desktops, and done quite well. But for those who take a much more conservative approach, I do think we're within a few years of reaching a point where virtualization, related tools, and processes will demonstrate a capacity to do more with less even in production environments.

I'm not sure that we'll see another explosion of growth in VMs like we've already witnessed... and for those players who came late to the game, oh well. I also believe there is still huge potential, but that progress from here will come at more manageable pace.

[Shibin Zhang]

My 2 cents:

1. Server cost is not the most important data center element. There are others VM can help.

2. Not all the servers are suitable for VM. I.e. servers running at high CPU (at peak time???) and servers providing real-time responses.

[Shibin Zhang]

I did some self-study of VM before, but I stopped since I thought I would never have a chance working on VM. Otherwise, I might have more comments.

[M Houlihan]

When you say 7% of all x86 have been virtualised you are assuming, by multiplying by 10 to get 70 virtuals, that they were converted to Hosts. I think the 7% figure means they have been P2V'd.

So in the instance of 100 servers. If 7 have been virtualised then there is 93 remaining. Since you need 1 host to support the 7 P2V'd servers you are left with 94 servers not 163.

Anyway that is how I interpret those figures.

Ciao

[Shibin Zhang]

After visiting many Blog sites (Dave’s, Chuck’s and so on), I really feel like this Web2.0 tool. I feel Blog not only provides a new way for people exchanging ideas and making fun, but also allows people to help each other. You help me out (i.e. providing solutions and suggestions) today and I will help you out tomorrow. As a result, everybody wins.

[Chad Sakac]

Dave, as always an insightful post.

My view, and my experiences with customers of all shapes and sizes: Server virtualization is just starting.

1) most customers have attacked the virtualization of "early hit" hosts (characterized by light cpu/memory loads), and in some cases are fearful of virtualizing mission-critical or heavy IO VM workloads. These work great of course (only exception being the real-time workloads that another poster noted), and we've done a ton of work on validating these apps (Exchange, SQL Server, Oracle, SAP, Sharepoint, etc...) at all sorts of workloads and VI3 configurations - but sometimes it just takes time.

2) While there are customers leveraging every advanced server virtualization feature under the sun - most are still realtively basic. I'm always amazed to find customer who have ELAs only using VM HA, occassionally VMotion, but rarely DRS. There is a wide adoption curve in front of us here beyond basic consolidation via a hypervisor. While VMware has the best hypervistor - more to the point - they have a wide and growing lead on advanced use cases and features.

3) Lack of operational discipline once virtualization is widely used is a real problem. When creating a server takes a click and a few seconds, you end up with loads of servers. This has management, change control, backup, and security implications that we're all working on building solutions for. Frankly, I think this is the "undiscovered territory" of this huge change.

More and more customers having seen a big return (personally, my lab has more than 450 ESX servers and 40 VC hosts + Lab Manager - we've saved millions, and across our company, there are thousands of ESX servers, and every advanced use case under the sun), are starting to ask the next question: "how do I apply the same phliosophy across my IO fabric, how do I unify that fabric, and the same approach to total storage virtualization" - we are starting to call this "Next Generation Datacenter (NGDC) Type 1"

Will be fun over the next 10 years!

[Tony Asaro]

Dave,

Natually I have been thinking about this a great deal. What we're finding is that server virtualization is about consolidation but more importantly about ease of management. Upgrades, implementing new apps, high availability - these things become like night and day for sys admins.

Another thing to consider - small and medium enterprises won't buy tons of physical servers - they will have 2-20 phys servers but now can have lots and lots of virtual servers.

[Mike]

Hmm. I tend to have seen very different approaches to virtualization. Sure some datacenters have had a very regimented P2V mindset that is easy to measure. Smaller orgs have taken physical servers with several SERVICES and given each a virtual server, or created backup or clustered virtual nodes. Developers and not-so-production environments have simply created as many VMs as their hardware can handle. Basically I see virtualization sort of defined differently in its use. Previously we could simply count physical nodes. Now I think we need to consider actual hosts, production hosts, services per server, and so on. Not exactly an apples to apples count of conversions...

[Nick Bernstein]

I think this is true: we're in the beginning of the trend, but I don't think it's the beginning overall -- what it looks like to me, is actually a resurfacing of the mainframe, but a mainframe made of open, distinct modular components including san/nas, blade servers, and virtualization.

[Ed]

I'm happy to see someone already noted certain workloads don't make sense to virtualize, although that's mitigated some by hardware designed specifically for virtualization. However, there's also certain workloads which fundamentally cannot be virtualized: if you virtualize your security software, it may be more secure than running all of that software on the same box without virtualization, but it is less secure than running that on a separate box for each service. For example, if you have a three-layer perimeter, made up of a firewall box, a proxy server box, and another firewall box, putting those as virtual systems on the same hardware is an incorrect choice, from a security perspective. (Of course, if you don't need the three layer perimeter, having hardware for it is also a security mistake.)

Also, I don't believe the numbers are being reported consistently. When some companies report they've virtualized 7% of their systems, they mean that of 100 servers, 7 of them are now virtual systems running off of 1 box - as stated above: 94 boxes, with 100 total systems.

However, others mean that 3% of their systems are virtual, running off the same box, which has grown additional virtual systems as well: 98 boxes, with 103 total systems.

Still others, bad at math, mean that 3% of their systems are virtual, running off the same box, which has grown additional virtual systems: 98 boxes, with 112 total systems.

Alternatively, still others, bad at math, mean that 3% of their systems are now virtual, running off the same box, which has grown additional virtual systems: 98 boxes, with 102 total systems.

Only the companies offering virtualization software seem to think anybody would be reporting it as you seemed to think above: 100 boxes, with 163 total systems. That may be how technical people talk about it some of the time, but when management speaks to reporters, they're going to be claiming the biggest virtualization number they think they can get away with.

Which, of course, leads to the dishonest: 7% virtual means 98 boxes, with 100 total systems. Or, for that matter, 94 boxes, with 96 total systems. (Why anyone would claim decommissioned servers as virtualized is beyond me, but I've heard of people doing crazier things.)

[Matt Booth]

Dave,
I hate to be counter, but I have to really question NetApps commitment to virtualization, given my recent experiences with your company and your IBM partnership. After nearly 6 months of struggling mightily with an N-series 5300, which I believe corresponds to a FAS3000, we have finally been told by a representative of your organization, that, rather than spending 1 minute attempting to resolve the issue, ESX3i is not in your operational matrix and they actively refuse to support it. As 3.5i has been on the market since January of this year and you are an active partner with VMWare, I would assume that your organization would be eager to support such an active and growing product line. According to a female in your support department, who apparently refused to be identified, not so. Even though a link from your "NETApp Interoperability" site references a VMWare support document that explicitly details support for this configuration. So as of now, unless you can offer any assistance, as this is directly affecting our customer base, I will be forced to recommend removing the NETApp products from service and replace them with clearly inferior products from EMC or God forbid Equalogix, most likely losing my job in the process, since I have been to this point an ardent NETApp/IBM defender/evangelist. Please feel free to contact me at any time regarding this issue.
Regards
Matt