IDC’s newest survey of server vendors shows boxes running every kind of operating system — except Unix — sold more during the last three months of 2010. Sales of Linux servers rose 29 percent; Windows rose 16.8 percent, but most surprisingly, sales of mainframes shot up 69 percent – the highest growth rate IDC ever found on mainframes.
In some cases the server vendor owns the OS (IBM & Oracle). How much can we rely on the accuracy of the numbers they report?
Does this take into account systems sold with no OS? Many large companies would have site licenses / support contracts and would not necessarily order the server with an OS installed. Others may be using freely available Linux distro’s, a BSD or something else.
Does anyone have any idea how IDC account for the above?
Good luck installing a free distro on a mainframe or even a big workstation. Even when possible, like for ProLiant/HP Zseries or PowerPC/SPARC servers, you lose your warrenty if you do that, you can’t have support when you need it. This is just not an option for most managers buying these things.
You don’t lose your warranty. At best you might lose some OS support from the vendor, if they even have that, but anyone going this route would get support from the community, hiring a company with techs for that particular OS or having the support in-house.
You lose driver support too, and its the main problem
As I am working in an international company, we are buying HW (big iron). IT/IS staff install themself and support, any OS required : Windows, Linux, BSD, ESX, etc.
And I really wonder if IDC does take this situation into account.
Here we have a lot of big irons, all of them running Debian.
We still have warranty on hardware.
The original poster of this line is deluded. IBM is the predominate mainframe maker in the world. AND they sell Linux on their mainframes. I know they support Red Hat but you might be able to get Suse also.
It is interesting to speak with sysadmins about the Mainframes running Linux: they say it really sucks. That Linux is unstable and they encounter all sorts of problem.
On top of that, a mainframe is not that fast, cpu wise. Any highend x86 cpu is at least twice as fast. In fact, a better number would be 5x faster. So you need 5 Mainframe cpus to match one high end x86 cpu. This is true, and I can provide links and articles that proves this fact.
MAinframe cpus are good at RAS. That reliability costs a fortune. They are very reliable and tailor made for reliability. But they are not fast.
x86 are very fast, but very unstable and buggy. Every once in a while, intel and AMD has to withdraw x86 cpus because of bugs. Mainframe cpus are not buggy, but they are slow.
Total BS. IBM’s z196 is a 5.2 Ghz, 4 core processor. The last time I looked it was the world’s fastest processor.
How do you know it is the world’s fastest cpu? Because IBM said so?
http://www.tested.com/news/ibm-claims-to-have-produced-worlds-faste…
And you trust IBM? Is it so?
Or do you have any evidence it is the world’s fastest cpu? Benchmarks? There are no benchmarks, because IBM won’t publish benchmarks. Why? Because it is very slow in reality. You want to see some links and other proof that show how slow the z196 “the worlds fastest cpu” actually is? Then you can show me links that show it is the worlds fastest cpu (there are none)? Deal?
Yeah, I’m curious as to how Linux on the mainframe evolves in the future, since the systems really are that different.
There’s really two reasons to be using a mainframe these days, and cpu power certainly isn’t one of them. They really are designed for two things these days – massive database loads, and reliability. There’s a reason a good percent of banks and insurance companies have mainframes, they have ridiculous levels of i/o compared to your average system.
As far as reliability, not only are the processors mirrored into a form of raid-1, with the z196 they’ve introduced raid for ram as well. Then of course you use parallel sysplex to mirror a few entire systems. There’s always a market for reliability like that. It may not be large, but until x86 systems start offering similar facilities the mainframe won’t be going away.
I doubt x86 will offer similar reliability, as x86 is too buggy and bloated. To get reliability, you need Mainframes / SPARC / POWER / Itanium. For instance, some SPARC cpus can rollback and replay instructions if something went wrong, just like Mainframes. Such functionality does not help performance, and it cost a fortune to implement.
Reliability (Mainframes) and performance (x86) are contradictory, pick one of them.
I can pick third variable in this not-so-simple equation – cost. If you were to buy hardware (with or without software) you have to be perfectly aware what’s going to run on top of it. If you are not sure, just buy a Dell appliance, for example, R710, and install whatever you wish on it – Red Hat, Novell suite, MS or even Solaris. That’s a good deal in terms of price. Sun and IBM have nothing to offer against it, again price-wise.
Bullshit.
We bought hundreds of HP and IBM servers and workstation with no OS pre-installed and used our RedHat and Microsoft site licenses to install them – and we never, and let me make it absolutely clear, NEVER, had any issues with warranty.
I greatly doubt that big companies use pre-installed as opposed to rolling using their own server/workstation/desktop images – be that for Windows or Linux.
– Gilboa
Edited 2011-03-05 09:55 UTC
You have to take figures from the likes of IDC and Gartner with a pinch of salt, due to the methodology they use to track installed operating systems. Most users do not order their operating system pre-installed, but go for none at all and either install Linux or Windows from an existing VLK. Either way, people like IDC have no way to measure that.
That would be a whole different situation if they factored in the consumer level systems (OSX, iOS, Android, etc)
The consumer level systems which you mentioned (OSX, iOS, Android) are not servers.
Hence it seems quite understandable that they are inot shown in “IDC’s newest survey of server vendors”.
Edited 2011-03-04 02:50 UTC
I got that … I was simply noting that if they had considered iOS, OSX, Android in that equation UNIX would surely be on top.
Linux isn’t Unix, it is rather a written-from-scratch independent implementation of the POSIX specification which has not been certified as POSIX compliant because it is expensive to do so.
Therefore Linux is just Linux. Consequently, unlike OSX (which was based on BSD code) and arguably perhaps iOS as well, Android cannot be counted as any kind of “Unix”.
Even if OSX and iOS machines were to be lumped in with the Unix figures, Android should be lumped in with the Linux figures.
Edited 2011-03-04 03:24 UTC
Support for this:
http://en.wikipedia.org/wiki/File:Unix_history.svg
Mac OSX comes from NextStep code which in turn comes from BSD code which in turn comes from Unix code.
GNU and GNU/Linux code spring up out of nowhere (no predecessor code for either of them). They were both written from scratch. GNU source code was originally written from scratch using editors on Unix systems, whereas Linux source code was written from scratch using the editor on a Minix system. Both are now supported on Linux systems.
Android derives from the Linux kernel code.
More information here:
http://en.wikipedia.org/wiki/Linux
Information about POSIX:
http://en.wikipedia.org/wiki/Posix
Note that even though Linux largely is POSIX compliant, Linux is not certified as POSIX compliant.
Edited 2011-03-04 06:01 UTC
Actually, OS X and the current BSDs do not contain any UNIX code.
All of them can be traced to the short-lived 386BSD. FreeBSD and NetBSD were directly based to 386BSD, OpenBSD is a NetBSD fork, DragonflyBSD is a FreeBSD fork and PC-BSD is to FreeBSD what Ubuntu is to Debian (sans the stupid flame wars). XNU’s (OS X’s kernel) BSD parts are also based on FreeBSD.
386BSD was a x86 port of the Net/2 release which was essentially a rewrite of all 4.3BSD code that was under AT&T’s copyright. So, while Net/2 and 4.3BSD were pretty much the same thing, there was no UNIX code in the former.
Yes Linux was written from scratch, so is Solaris or AIX or HP-UX or even *BSD. All of them are Unix!
No, Solaris or AIX or HP-UX and even *BSD were not written from scratch. They all inherited an existing codebase that came before them, they are all “forks” of an earlier version of Unix. They are all Unix.
Only Linux, and the original Unix itself, were written from scratch.
Linux is not Unix. There is not one line of Unix source code in Linux. Linux was written from scratch, by Linus Torvalds.
Like so:
http://en.wikipedia.org/wiki/File:Unix_history.svg
In this figure, solid lines entering from the left represent existing source code.
Do you understand now?
PS: BTW, the GNU/Linux OS uses GNU (and GNU was written from scratch, by Richard Stallman), but the Linux kernel itself does not use GNU, so the Linux kernel was also written from scratch. Android uses the Linux kernel, but it does not use GNU AFAIK.
GNU is also not Unix. In fact, the very abbreviation, GNU, satnds for GNU is Not Unix.
http://en.wikipedia.org/wiki/GNU
“GNU is a recursive acronym for “GNU’s Not Unix!”, chosen because GNU’s design is Unix-like, but differs from Unix by being free software and containing no Unix code.”
Edited 2011-03-04 12:27 UTC
Linux is not Unix. Period.
It’s not based on Unix code, it’s not Unix certified. It’s not Unix.
AIX, HP-UX and BSD all share a common heritage which Linux does not.
In fact, Linux wasn’t even intended to be a Unix clone. Originally it was a Minix clone (which also is not Unix).
So anyone who tells you that Linux is any way related to Unix (beyond architecture similarities) is simply WRONG.
Linux was not intended to be a clone of Minix. Minix is a microkernel design, Linux is not, it is a monolithic kernel. They are fundamentally different designs.
http://en.wikipedia.org/wiki/Minix#MINIX_and_Linux
This more detailed picture purports to show the Evolution of Unix and Unix-like systems:
http://upload.wikimedia.org/wikipedia/commons/5/50/Unix_history-sim…
Minux and Linux are shown on the left, neither of them inherit any code from any lines of Unix development, nor from each other.
Minux and Linux are Unix-like systems, but they are NOT Unix.
Minix is Minix.
Linux is Linux.
Neither is Unix, or derived from Unix.
Edited 2011-03-04 14:25 UTC
You misunderstand me. I know Minix and Linux are entirely separate entities, but I thought I read that Linux was originally built to be a Minix clone rather than Unix (can’t remember where I read that though)
After going back and reading wikipedia, I suspect I read something like the following and got my wires crossed:
I’m definitely under no illusion as to what Minix, Linux and Unix currently are nor where the code base originated from.
Linux is a clean-room clone of Minix. There was never any UNIX code in Linux. It’s is the bastard step-child of UNIX’ distant cousin Minix. No actual blood relation; no actual code history.
BSD is a continuation of AT&T UNIX source code, with a bunch of extra features added in. By the year 1992, there was no original AT&T UNIX code in BSD anymore, but the history is still there, and you can trace back through the CVS commit logs back to the original UNIX sources.
While none of the free implementations of BSD have paid the exorbitant fees to go through the certification process and thus none can call themselves UNIX(tm), they are most definitely Unix-based systems.
MacOS X is a crazy hibrid of NeXT, FreeBSD, NetBSD, and Mach, giving it direct links into the UNIX source code lineage. And, Apple has paid for specific versions to go through the certification process. Thus, it is also UNIX(tm).
But there is no way, no how, none possible, for Linux to be considered UNIX(tm), or even Unix-based, or even Unix-compatible.
No, Linux is not Unix. It does not follow the Unix specification.
This was a study about server installations, not unix.
No UNIX wouldn’t as Android is Linux and Android outsells iOS.
Plus your point is completely irrelevant anyway as this discussion is about servers not mobile phones.
I mean, if we really want to count irrelevant figures, then why aren’t router firmware and sat navs included either. In fact, I’m pretty sure my DVD player runs an OS too. :p
Edited 2011-03-04 08:36 UTC
This shouldn’t really be surprising to anyone. UNIX vendors themselves have been increasingly supporting Linux based solutions at the low end and reserving the likes of AIX and Solaris to the bigger server installations. Sun was the only one of the major UNIX vendors to show even the slightest interest in the small business sector and even that was a cluster****.
OSX is UNIX.
http://arstechnica.com/apple/news/2007/08/mac-os-x-leopard-receives…
http://www.opengroup.org/openbrand/certificates/1190p.pdf
Agreed, OSX is derived from BSD Unix.
However the Android kernel is Linux, and that is not derived from any Unix code at all.
Besides, neither OSX, nor iOS, nor Android are server OSes.
Mac OS X is a server operating system, it is used and shipped on Apple’s Xserve rack-units.
… which have such a small share they make impact on the numbers.
But of course that has nothing to do whether OS X is a server operating system, it can well-handle the tasks that would be delegated to a server. It even had some rack-hardware produced by Apple, for a time.
True.
However, Mac OSX is only a minority bit player in the OS market, and for this reason (in the popular sense) Mac OSX is not considered one of the mainstream server OSes.
But this will change with the shipping of Lion given they no longer differentiate between server and desktop by having two distinct products. With that change in how their products are lined up will raise the question whether the way server shipments by way of operating system is particularly reliable given that what ships doesn’t necessarily point to the said device being used as an actual server or an actual desktop – case in point, a Mac Pro can be a desktop but in the future when Lion ships it can be a server with the appropriate packages added after its shipped.
This is currently already true with Linux and for example Windows 2000.
True, but with Windows there is a 10 connection limit.
Only on fileshares, or IIS, if you installed say apache, on Windows 2000 WS, there would be no limit.
Yes, in theory but I’m referring to packages that come with the distribution itself rather than sourcing third parties. That put aside for a moment, IIRC isn’t that a licence violation? Aren’t there limitations placed on throughput within the registry in the case of the TCP/IP settings etc? I’m not particularly clued up on Windows so if you can provide greater insight it would be much appreciated.
Edited 2011-03-07 02:53 UTC
Not really, you get no support from MS, but they really can’t stop you.