Windows Bumps Unix as Top Server OS

Ivan writes " Windows narrowly bumped Unix in 2005 to claim the top spot in server sales for the first time, according to a new report from IDC. Computer makers sold $17.7 billion worth of Windows servers worldwide in 2005 compared with $17.5 billion in Unix servers, IDC analyst Matthew Eastwood said of the firm's latest Server Tracker market share report. "It's the first time Unix was not top overall since before the Tracker started in 1996.""

Comments

How long

[EraserMouseMan]

do you think it will last? Is Windows picking up momentum or is Unix losing momentum?

Re:How long

[Lumpy]

How about the 12 servers we bought from dell without an OS that had linux installed on them? I know those were not counted. Or the 100+ servers acting as mpeg2 video routers in the headends that were also bought without an OS that has linux installed as well.

I bet the number of intel based servers without an OS sold is far greater than the number sold with an OS.

Unless they release ALL the data it's a worthless study. have dell release all server sales with and without OS. without OS will dominate nearly 2 to 1.

Re:How long

[fitten]

This is why Windows/Linux eats into HP-UX/AIX/Solaris market share.

Years ago we knew that the first casualty of Linux would be the proprietary Unix companies. The workstations first and then the servers. Although Linux is advocated as a Windows replacement most of the time, it's an even better Unix replacement. As Linux improves, it will just hurt Unix more. A friend works at a place where they've replaced almost all their Sun servers with Linux servers except the cluster of V880s that they have to still run certain software packages (Solaris only binaries). I could easily see them replacing those boxes with multi-cpu/core Opteron boxes (maybe even from Sun) running Linux if they had that software available. This is a place that has purchased multiple Sun E10Ks and multiple SGI O2Ks and the like in the past. Now, they are mostly Linux except where they have entrenched software or have issues where they need large systems (32p and 64p) and Linux doesn't work on them for some reason or work well on them.

Re:How long

[Zeinfeld]

I agree that Linux would have come higher if the eventual OS installs of servers with no OS installs had been recorded, though.

The researchers claim to have adjusted for that effect. Most servers are sold without an O/S because even machines bought as Windows boxes are likely to have the O/S loaded under a site license.

The non-Linux market for UNIX continues to shrink. As it does machines move from expensive proprietary platforms like HP, Sun or AIX to commodity Intel/AMD boxes. This means that Linux is effectively handicapped against the traditional UNIX varieties in this race, as is Microsoft of course.

Servers have been getting cheaper for years, the server market changes as a result. Forty years ago servers were mostly $1 million plus mainframes. Today its a definition thing, you can buy a 'server' for $100 in Frys and hang a printer off it.

All the growth in the market comes at the bottom end as small businesses start to invest in infrastructure. A law office with ten employees using Windows XP is going to buy a Windows server, end of story. An ISP with 100 Linux boxes doing hosted web is going to buy Linux for machines 101, 102,...

I don't think the survey is actually measuring real transitions. There is no compelling reason to move from Linux to Windows or from Windows to Linux if you have installed base. There is a huge cost incentive to move from proprietary UNIX to Linux. There is also a major incentive to introduce Windows Server systems to provide support infrastructure for networks of Windows machines.

There are relatively few areas of overlap between Windows and Linux. Both can host Web sites, but once you have developed active server pages you are locked into Windows. Both can host a mail server, but people do not buy Exchange as just a mail server, the calendar features are the real value.

Sheer number of small servers

[PIPBoy3000]

This probably reflects the massive number of smaller servers that are out there, which often have Windows installed. In our organization, Windows servers tend to have a single application on them (typically by request of the vendor), while our Unix and AS/400 servers tend to have dozens of applications on them.

The irony is that Windows applications often "don't play well together", making it almost a requirement that they get a dedicated piece of hardware. As a reward for this problem, their rankings are boosted.

Re:Sheer number of small servers

[Flying pig]

one word: VMWare

two acronyms: ESX,GSX

that should start to get the numbers down.

Re:Sheer number of small servers

[AKAImBatman]

However, we _also_ have two Linux servers. One of them, is the main mail server and the other is the cvs repository.

However, if you were running these on Big Iron Unix machines you'd have them both on the same server. There's no point in wasting the resources of a large machine on a single task.

I should think that the number of "small" Linux boxes is now pretty close to the number of "small" Windows boxes.... but I doubt whether they come into these stats.

Similarly, what about all the routers running Linux variants. Do they count? They're "servers" of a kind.

Unfortunately, none of the Linux boxes count for much. If you purchased an Enterprise license, then they might. But every system that has a downloaded copy of the software, free software w/ paid support, or bundled Linux customized for the box, most likely does not count toward the revenue totals. And even if it does, such copies would often be valued lower than their Windows counterparts due to a lack of standard pricing on "free" software.

It sucks, but that's the way these studies work. The only way to find somewhat accurate numbers is to poll a significant number of businesses (100 or more) for an exact count of machines by OS.

In nature mono-cultures like that would die-out very quicky.

It happened to the American Indians. Their ability to resist disease was limited to only a few types. (You'll forgive me, but I forget the exact terminology used here.) When the early explorers arrived they brought new diseases with them (especially in their pigs) that most of the Indian population was unable to resist. Thus the great cities that existed in the time of the Conquistadors had completely disappeared by the time the Mayflower settlers landed.

Unix != Linux

[AntiDragon]

The numbers - they make me sleepy...

But note that the article mentions the growth of both Linux _and_ Windows. This is really about the ongoing decline of pure UNIX mainfarmes - something we've all been aware of for years.

The fact that Windows OS now outnumbers UNIX boxes is neither suprising nor noteworthy. They've been chipping away at the server market for ages. Bound to happen eventually.

But what I would be more interested in is out of all these switchers, what's the ratio that switch to Linux compared to Windows? Linux growth is faster (Upgrades along the Windows path don't count, we're talking complete platform migration) I believe. But naturally the title of the article gives enough bias to encourage readers to miss that little tidbit. Or maybe using the phrase "Windows beats Unix" is the journalistic equivalent of shouting "Fire!" when it comes to grabbbing attention... :D

Re:I have some numbers...

[jimicus]

We purchased five brand new Dell rackmountable servers last month. When we got them, we burned in some linux and threw the windows disks in the trash...

Seeing as Dell doesn't force you to buy an operating system with their servers, why did you bother buying them in the first place?

Re:Linux?

[bubulubugoth]

Simple, Linux isnt cheap for opterons, powers, and "enterprise class distributions/support".

SuSe for OpenPower costs about 800 usd.

Redhat AES, the same, BUT, redhat charges you by installation, so, virtualization is more expensive with Redhat than SuSe.

Also, since IBMs OpenPower Machines, only runs on linux, they eat a chunk of market to aix (pseries), and even more, HPs Intanium and Opteron, run with linux, add this redhat/suse sales from Dell, and you will have a very and rich environment...

You still can always get a debian, yellowdog or something weirder to run on HP Opteron or IBM OpenPower, but why loose guarantee for a 25k-30k range machinery?

Free servers

[jamesl]

Maybe its because Sun is giving away servers. For free. No cost. And each free server would add ... let me think ... ummm ... zero dollars to the total.
http://blogs.sun.com/roller/page/jonathan/20051218 /

Maybe not.

EU Competition Authority

[bfree]

Doesn't this help the EU Competition Authority to argue that Microsoft is actively extending their monopoly on desktops into the server market? Does it therefore also suggest that for once a "government" is acting on something in time, saving a market from an extending monopoly before the monopoly covers the second market? It doesn't do anything to make Microsoft comply with court orders though.

Re:not necessarily

[Fished]

I have a Solaris web server with an uptime of 2436 days. It's last outage was when we moved it from one location to another. 'nuff said.

(Yes, it should have been patched, etc., but as it turns out this server is running Solaris 2.5.1, and everyone forgot it was there. the amazing thing is that it has run for over 6 YEARS without a reboot.)

Or UNIX got free...

[cloricus]

When it came time to look into getting more UNIX based servers at work we took a new path over RedHat/AIX/SCO and settled on GNU Debian Linux. We are now running three new servers (in the 2005 year) using UNIX(-like) on top of our existing UNIX/Windows setup.

So really sales figures can no longer really be an indicator of what is really out there now that businesses are happy to buy blank servers and load their favourite Linux/BSD distro...unless many corporation's are running pirated versions of Windows on their servers. Which I for one would seriously doubt.

Re:Servers

[ajs318]

..... Which is why I make a point, anytime I am forced to buy software I will not be using, of "deregistering" my purchase. I send a letter to the vendor stating in no uncertain terms that I do NOT accept the EULA offered with the software and that I will consider my rights violated, with the Usual Consequences, if I am counted amongst its registered users.

Re:Unix servers

[Medievalist]

As a system administrator who has recently spent time trying to mitigate the security consequences of using NFS for a large campus network, I wonder if Windows may not have some advantages.

I feel for you, but I've had extensive experience in both, and I can tell you that SMB/CIFS is just as bad as NFS.

NFS security is Unix security writ large and networked: if it's not root it's not important. If your machine has the right IP, and you've got root on the box, switch your UID and NFS gives you all priviliges for that user.

That's true. However, effectively Windows Networking is the same (effectively, I said - I know the details are quite different). If a user has full administrative rights on a Windows system that is a full domain member then network file storage as a whole is easily cracked; in fact you can probably easily derive everyone's cleartext passwords by requesting them in a weakly encrypted form and running them through some free software. In every networked file system I know of, if you have unrestricted access to a node that has full membership in the authentication protocol, you have the same ability. Some systems make this harder, by restricting the nodes that have full authentication capabilities (Kerberos comes to mind immediately) but again, if you have root on the Kerberos server or SASL store or whatever, you can get passwords, and if you p0wn a trusted multiuser machine you either exhaustively query the authentication system or stealthily pick up the user authentication converstations as they go by.

And NFS is the ubiquitous Unix Network Filesystem!

No. It seems that way if you hang out in "traditional academia" type shops but really NFS is merely popular, not ubiquitous. I have worked in computer science for decades in defense, research, industry, and finance, and most of the places I have worked had a total prohibition on NFS because of the inherent insecurity of the early implementations.

Goddamn, what a security mess. I'm looking at alternatives like OpenAFS or Coda -- but hell, those aren't very mainstream.

I think they are essentially immature, but fundamentally better in basic design. They do not make the assumptions that NFS makes (NFS works best with NIS or LDAP to co-ordinate user IDs, but NIS has horrible design flaws and LDAP is also immature despite fairly strong design). Tridge recommended Coda to me when I asked him about alternatives to NFS and SMB years ago, and I respect his opinion rather highly (we should all probably move to Coda or Andrew so that the implementations can mature!)

I could just use samba for everything I guess.

There are great advantages and small disadvantages to that approach. But you need LDAP to make it really integrate properly, and the samba team will always be facing challenges from Microsoft's evolving use of crypto that will make their product lag slightly behind Microsofts' implementation releases (granted, once the samba stuff catches up it's more robust and scaleable).

But why not go all the way and run Active Directory servers?

Because it's less secure than NFSv4 and LDAP, and far less cost-effective than Samba, and less compatible with non-microsoft systems than OpenLDAP or Red Hat directory server. If you have a highly skilled computer staff and relatively unskilled end-users (a situation often found in a profitable corporation) samba and OpenLDAP are a nice combination. Don't try it if your IS group is a bunch of talentless hacks, though!

Re:Virtual Server

[digidave]

Why virtualize?

1. VMWare makes backups much easier. Just compress and copy a directory.

2. Disaster recovery. Did your main server just go down with five VMWare guests? No problem, just copy your recent backups (or from the crashed server's hard drive if it still works) to a new server's VMWare installation. No setting up all those apps and OS configurations. The VMWare host is a very simple installation that is easy to recover since no non-default apps other than VMWare are needed.

3. Do more with less hardware. Many companies try not to buy hardware unless it's abolutely necessary. It's great to be able to create a new development or testing server at the drop of a hat without needing new hardware.

TV Killed the Radio Star

[Gary W. Longsine]

Years ago we knew that the first casualty of Linux would be the proprietary Unix companies. The workstations first and then the servers.

This seems to be conventional wisdom around these parts, but it's not backed up by evidence. The UNIX vendors that have died to date have nearly all been killed by inept management, including the next one expected to kick off any day now, SGI.

Of course, they had some assistance from early Windows marketing hype and a lazy trade press that believed that Windows would take over the server market in 1992 or 1994 and continued to believe it for over a decade despite overwhelming evidence that the product wasn't ready for the enterprise server room.

And Linux has been taking over the UNIX workstation market? Give me a break. That market has been dead for almost ten years. Windows took over the market niche formerly occupied by UNIX workstations (including X-Windows stations which were not full UNIX boxen) long before Linux was ready, and the market niche doesn't really exist any longer -- it became part of "the Windows Desktop".

Although Linux hasn't killed off any UNIX vendors yet, they appear to be concerned by the possibility. IBM for example has been perfecting their AIX up-selling technique -- hook clients with Linux advertising, then up-sell them to AIX. They have a different term for it, migration analysis or something, which they do free for their customers. (Apparently it works well enough that one IBM group pays cash money to another IBM group to do it, such that the customers don't need to pay for the proposal, which says something like, "Gosh, who wouldda thunk? It turns out that your situation lends itself to an AIX solution after all. Shucks, it's a good thing we did this study or you would have been migrating to Linux and you wouldn't be able to leverage the AIX value proposition" or something like that.) IBM is also hedging its bets by making some more serious investments in Linux, and trying to create a market for Linux on IBM hardware, both Intel and Power based.

Linux has been making inroads into the server market (as you illustrate by example) but it hasn't killed a UNIX vendor there yet. It's also making hay in the embedded systems market. In the process it is displacing some UNIX and some Windows, but also (and perhaps mainly thus far) growing into new areas where there were no dominant players (network linkup boxes were simpletons until fairly recently and didn't run a full operating system like modern switches do, for example). That didn't kill any UNIX vendors, either.

Windows isn't a stationary target, of course. The expected growth of the product in the server market is finally happening, albeit ten years after the fact. This means the market thinks that Windows is an acceptable substitute for many of the former UNIX server tasks. Even if UNIX administrators have plenty of good reasons why it's not, clearly the show stopping problems which prevented its rise for the last ten years are behind it.

The frame of reference seems even to have a waning validity. At the very least, analyzing the question for the past was fairly simple, but it becomes very much more complicated to analyze contemporary events through this lens, since most of the surviving UNIX vendors are also Linux vendors. Things have changed so much in the last several years that events won't make sense when viewed through this lens at all. Allow me to illustrate the problem:

SGI probably sells more Linux than IRIX at this point. If and when SGI hacks up the last bloody phlem and finally dies, which of the following will have occurred?

1. [ ] Windows killed a UNIX vendor
2. [ ] Linux killed a UNIX vendor
3. [ ] management ineptitude killed a UNIX vendor
4. [ ] Windows killed a Linux vendor

Hint: All of the above.