Microsoft Cheaper For Web Serving?

Tinman_au asks: "Bink.nu has an article titled "Leading Belgian Hosting Provider Realizes Lower TCO on Windows than Linux" that asks the following: 'Many total cost of ownership (TCO) studies have reaffirmed that TCO of a large enterprise infrastructure based on Microsoft Windows Server 2003 is lower than one based on Linux. But what about TCO in a Web hosting environment?' In the table of figures, the cost area breakout lists labour for Fedora at 77.88% with Windows .NET with SQL Server 2005 as only 53.15%. Admittedly, the report was done by Microsoft itself, so I guess it couldn't exactly be considered impartial, but not being a web admin I found myself wondering, is Windows really that much easier to look after in a web server environment, or has Microsoft fudged some numbers?"

Comments

I'm going to have to ahead

[antifoidulus]

and tag this one "flamebait"

Re:I'm going to have to ahead

[arivanov]

Disagree. I will tag this as "Logical Development".

This study perfectly describes the problem with many Linux/Unix deployments out there. They are done by people who take the approach which they have grown accustomed to on Windows, Novell and the like and try to transfer it to Linux/Unix. This approach is best described as "everything you cannot do with the vendor tools must be done manually" and "we only use commercial/vendor software". When using this approach Linux/Unix invariably results in higher TCO because the price of labour is higher and level of one-click moron-friendly automation is lower for most cases.

When doing Linux/Unix work writing your own tools and assisting yourself in automating tasks is a part of the job and Sysadmins who do not possess the skills should not allow themselves to claim that they are Linux/Unix Sysadmins. From there on, if you estimate the costs of running and deploying systems without taking this into account you invariable come up with Windows being cheaper.

That is the reality, face it move along and ignore the study. While it was using the right analysis methods it was analysing deployments which do not use the correct design and process for either system. If you use design and processes which are wrong for one system it is not particularly surprising that you get bad TCO for it.

Re:I'm going to have to ahead

[CastrTroy]

I think the reason that you can say the TCO of windows is lower is because you can pay some brain dead monkey to be a sysadmin, and have it work. It won't be well optimized, and it won't be completely secure, but for the most part it will work. Now consider Linux. You can't just really pay some guy with an MCDBA/MC??? or equivalent to operate your systems, because there isn't really any equivalent of that in the Linux world. Even most self taught Linux people are more knowledgeable than a lot of the "Microsoft Certified" people out there. So, because the Linux tech actually has more talent, and actually deserves to get paid more. It's like comparing the average Perl programmer to the average .Net programmer. On average, the Perl programmers will be more skilled and therefore more expensive. There's a lot of .Net programmers who took some 4 month course, and think they are now programmers, and a lot of companies hire these people. Studies show that .Net developers cost less, because they are less qualified. Not to say all .Net developers are idiots, but working in .Net development myself, I have to say there's quite a few of them.

Re:I'm going to have to ahead

[arivanov]

You are half the way there in your analysis.

Studies like this actually take into account the fact that the Unix/Linux sysadmin on the average is 20%+ more expensive.

They are wrong elsewhere.

They do not take into account that a selfrespecting Unix/Linux sysadmin will automate everything he/she can and will not repeat everyday mundane tasks. Instead of this they still count the time which is essential to maintain and patch the systems towards the TCO bill and multiply it by their number (correcting only for vendor tools to assist rollout where applicable). There is no correction for ad-hoc scripting and no correction for productising ad-hoc tools for internal rollout. Further to this many places go into the idiocy of prohibiting such internal software development. In fact I know one or two places where such activities are a sackable offence.

I have stopped counting how many times over the years I have heard the "We are not software developers" mantra from PHB wannabies. That is the damn difference between a high level Unix sysadmin and a Windows sysadmin in the first place. The Unix sysadmin can write in at least 2-3 rapid development languages - (k)shell, perl and/or python and the reason why he/she gets more money is exactly this. Paying him this money and not using this ability is stupid, but this is what many places do as a matter of policy. It is no wonder that places like this have better TCO under Windows compared to Unix/Linux. That is to be expected and that will continue to be the case until they start to automate mundane operations in-house, formally maintain the automation and productise/package it for internal use.

In fact the TCO numbers for systems like the one in the article (1000+ of slightly customized commodity software on commodity OS) come out in favour of Linux/Unix even if this activity is subcontracted out. They do not come out right only when it is prohibited and the work is done solely via vendor supplied tools.

Re:I'm going to have to ahead

[Foofoobar]

Speaking as someone who set up a LAMP architecture for a Microsoft vendor, my uptimes dwarfed those of IIS with the only time I had to take the server down was for software updates. The time it took me to do my makes and set everything up perfectly the way I want took a couple hours (not that different from Windows). When I left, they decided to switch everything over to an ALL WINDOWS environment. Not counting costs involved in switching all applications over, they had to spend almost $100,000 to duplicate the setup I had for free with open source. Costs in setup are minimal in LAMP, longterm savings and uptime are HUGE!! Also with LAMP, I'm a one man army. In my old company, they had to hire contractors for the DB work separate from the web dev. More labor, higher costs. With LAMP I can keep my costs very very low. This is yet again 100% FUD from Microsoft

Re:I'm going to have to ahead

[porkThreeWays]

I think you hit on a REALLY good point in there. In my *nix administration, I have no daily activities! I've written programs for them all. This is almost impossible to do for most GUI's in general and more so with windows. I've always found the task scheduler to be a very weak replacement for cron and windows event log to be a weak replacement for syslog. The funny thing is... I do actually have a report I have to run manually once a week, and it's for the single windows box I maintain.

Re:I'm going to have to ahead

[Sprinkels]

They do not take into account that a selfrespecting Unix/Linux sysadmin will automate everything he/she can and will not repeat everyday mundane tasks.

I see no reason why a selfrespecting Windows sysadmin would not do the same.

That is the damn difference between a high level Unix sysadmin and a Windows sysadmin in the first place.

Actually, to a high level sysadmin there isn't a lot of difference between administrating Windows and Unix like operating systems.

It's a prejudice to assume that scripting like on Unix is not possible on Windows. On the contrary, many Windows sysadmins use scripting tools to automate everyday mundane tasks. In fact even the archaic MS-DOS from the eighties has its own scripting language built in. Which is used even today by nearly all Windows sysadmins.

Re:I'm going to have to ahead

[walt-sjc]

I see no reason why a selfrespecting Windows sysadmin would not do the same.

Maybe because MS has gone out of their way to make automation Very difficult, instead pushing the pointy clicky interface for everything, and hiding all useful information in binary blobs inside the registry. Command line interfaces are klunky at best, and usually poorly documented. Yes, you can write applications to munge the registry, but it is a PITA. You don't bother writing automation applications unless you have a LOT of servers.

*nix automation is trivial. Small, simple scripts can do a lot. The information you need is right there, your script just parses what you see, and regurgitates what you would do manually.

The big difference is that the manual interface on Windows is gui, and the manual interface on *nix is command line (which is easy to script.)

Re:I'm going to have to ahead

[arivanov]

Exactly.

And as the subject of the article is TCO in terms of TCO customised per-site Windows automation costs much more then similar Unix automation. As a result it is worth it and justified financially only for very large installations. Everything else aside the tools (VB, ActivePerl and friends) cost money and you need a reasonable number of servers to get return on investment.

Custom Unix automation as means of reducing your TCO starts making sense from one server onwards. Tools are part of the platform and cost nothing.

So yes, selfrespecting windows sysadmins may automate custom tasks on Windows as well, but this for anything but very large sites may actually increase your TCO instead of reducing it due to cost of tools and cost of prototyping.

Typical

[techno_dan]

Yet again, another study show lower TCO on the windows platform. From years of Real experience I can say that this is not and will never be true, at least in the short term. The reason I say this, is that I can get more flexibility and horsepower out of a none MS deployment. This is not to say that MS products do not have their place, just that the studies are always narrow, and extremely limited in scope.

Re:Typical

[Spiked_Three]

You are saying the right thing, but not qualifying it correctly.

There is absolutely more horsepower and flexibility in the *nix environment, BUT; There is a steeper learning curve. Given 2 admin that are experts in each, *nix will give you more. Given 2 newbs, the Windows environment will get you up and running a medium complex web site faster, cheaper. The majority of enterprises work that way, newbs and short on staff.

What you're used to

[andy753421]

For me it's 10 times easier for me to fix up an httpd.conf or some .htaccess files and set some permissions with chmod/chgrp, but for other people using the IIS dialogs or whatever might be easier.

If I gave my grandma a IIS machine and a putty window SSH'd into an GNU/Linux/Apache box I'm guessing she'd get farther with the IIS machine, but on the other hand if you give those to seasoned veterans I would bet the apache box would be set up quicker.

No

[a16]

I work for a reasonably sized host (ie. thousands of unique clients, not 25 clients like your average host who will probably reply here), and as we are a completely linux (CentOS/Fedora) host, our operating system licensing costs are $0. If we were running Windows and SQL server etc, I'd estimate that our licensing costs per year would be 5-6 figure figures for commercial MS licenses for the number of servers that we have and the MS software that we'd need.

We have staff to administrate the servers, and we'd need them if were to manage windows servers. We generally only ever have 1-2 technicians available at any one time to manage all of our servers, and we'd need that many if we were managing the same number of Windows servers too. Ignoring start up training costs, which really only exist if you're migrating from Windows to Linux, staffing costs are absolutely no more for managing Linux boxes than Windows, I'd argue the opposite. Infact, if we were to migrate to Windows tomorrow, as TFA is saying we should - there would be huge initial licensing and training costs, I imagine more so than moving a Windows staff to Linux.

Sponsored by MS means this can be ignored, why do we keep posting this stuff? :)

Re:No

[WebHostingGuy]

That is why this TCO says Linux is higher. If you look at all the numbers Linux is lower on everything except Labor. They quote almost a 100% over Windows labor. Obviously this is because they can't find anyone capable of performing Linux system administration.

I wouldn't say this would translate to the rest of the world because your labor pool is going to vary from county to country, city to city and even among different companies. For this place in the world, at this time, this is probably correct. However, all you can extrapolate from this study is that for this place in the world, at this time, Windows is cheaper to run than Linux "for them." For everyone else you have to do your own TCO.

Re:No

[julesh]

They quote almost a 100% over Windows labor.

You have to wonder about the quality of their Windows admins who are willing to work for half of what Linux admins are.

'Cause you can't be telling me that a Windows server needs less maintenance time in the hands of sombody who knows what they're doing. I like to think I do, and I manage web servers running on each platform (Windows Server 2003 and Debian Sarge), and I'll tell you now I spend much less time on the Debian one. Updates are easier to apply, and there are fewer of them.

Re:No

[alexhs]

If you look at all the numbers Linux is lower on everything except Labor.You're not the first to fall for these meaningless values, but you're the highest moderated.

Percentages are meaningless if you don't take total cost in account.

Multiplying percentages by the given TCO you get (in eurocents per site/server/month):

(.Net 2000, .Net 2005, Linux)

Hardware 5.49 2.75 2.77
Network infrastructure 5.06 2.53 2.41
Operations and Network Mgt. 17.48 8.76 8.40
Power 1.69 0.84 0.85
Bandwidth 43.39 21.19 36.34
System software 12.90 6.45 3.31
Application software 50.02 28.64 22.75
Back office software 44.22 22.09 22.10
Labor 211.76 105.77 348.12
Downtime charges 0.00 0.00 0.00

But when you look at percentages, those for SQL 2000 and 2005 are quite similar. It means that one real server under SQL2005 hosts twice as much virtual servers as SQL2000.

I will let to others in-depth critics about the methodology.
Just that quote from the full report (emphasis mine): "Hostbasket experiences a lower TCO on Windows than Linux because our support cost for Windows is lower and because our developers and system engineers have better knowledge of Windows than Linux," notes Hostbasket Chief Operating Officer Alex Van Overloop.

Item-by-item comparison

[xoran99]

Isn't it strange that, in all categories but labor, the Linux solution was much cheaper? Why would it use less bandwidth? Why would the network infrastructure be cheaper?

In any case, I'm tired of TCO stories. Every last one of them is flamebait, and now I've read my last one.

Re:Item-by-item comparison

[Hijacked Public]

Why would it use less bandwidth?

Patch Tuesday?

Link to Report

[achillean]

Here's the direct link to the cost analysis report: TCO Report

Alas...

[Anonymous Coward]

I am, even as I type this, taking a break from trying to deploy a PHP app to a different server than the one it was developped on, and the amount of fiddling required to get both installations of PHP to work in a compatible way is mind-numbing. And costly. Turns out that PHP broke backwards compatibility again in its last version, which breaks the app, and the previous version against which the app was developped (5.1) has a security hole so that's a no go either.

I can heartily believe that in such a situation, a more industrialized solution (IIS+ASP or Apache+non-PHP) would be significantly cheaper to deploy.

Not gonna rant about PHP and beat that dying horse, but please don't dismiss such a study just because its findings annoys us. We have a lot of improvement to do still. :(

Re:Alas...

[Lalakis]

Turns out that PHP broke backwards compatibility again in its last version, which breaks the app, and the previous version against which the app was developped (5.1) has a security hole so that's a no go either.

2 things:
1) Chances are the app you are working with was badly written in the first place
2) You should have used a SUPPORTED distribution, appropriate for this kind of work. If for example you had used Redhat Enterprise Linux then redhat would have backported the patch for the security bug to your version and you would have a secure fully patched system without changing any software version.

Your company made some wrong choises and pays for them. It would be better if it had hired a qualified sysadmin in the first place and avoided all that... (and yes, he would cost more than a windows sysadmin)

Like everything, it depends

[Colin Smith]

The only area that Windows costs are cheaper in this study is "Labour".

Typically, your windows admin is a little cheaper. Typically apache can handle more virtual domains more reliably and requires fewer staff to manage. If you have 2 Windows staff and 2 Linux staff then Linux could be seen to be more expensive. The question is, do you really need 2 Linux staff, are they spending half their time idle? Are you using best infrastructure practice to manage your machines or are you installing each by hand? Are the Linux staff simply more senior within the organisation and therefore paid even higher? Or if you break it down by domain rather than by server, do the costs come out the same?

The study is deliberately oversimplified to hide the details of where the money's going. After all, it's propaganda.

Typical?

[Anonymous Brave Guy]

Ironically, I suspect your comment actually demonstrates why, in real terms, a lot of businesses find MS cheaper.

You may get more horsepower and flexibility out of a non-MS environment. That's great, and makes non-MS the way to go if a business is employing people like you.

Now, would you describe yourself (being honest) as a smarter-than-average sysadmin, a Linux/Mac/whatever specialist, an experienced geek...? In other words, are you a typical sysadmin that a typical company will hire, with typical experience on the various platforms, or would such a person require more experience/training/skill to get the same good results out of non-MS systems that you do?

On the flip side, do you (being honest) have less than average experience/skill with MS systems, perhaps as a result of specialising elsewhere, and would you therefore require more training and expertise to get the same quality of results others do out of MS server software?

Obviously, I can't read your mind, and I'm not going to put words into your fingertips by guessing your answers. But I can make an educated guess that there are a lot more people around who know how to get OKish results out of MS stuff than there are who know how to get much better results out of non-MS stuff, and that the MS-using folks therefore tend to be easier to find and cheaper to hire. That has a major effect on the bottom line of a business, and is why (for many places) MS is going to look like the safer bet on TCO grounds for at least a while yet.

Re:Sadly, I'd have to agree

[julesh]

I work for a charity, and we run Windows 2003 on our web servers.

I'd have to agree with the findings, simply because any idiot (like me) can run a Windows 2003 server. All you need is Windows experience (which everyone has nowadays). Linux requires special knowledge and/or training.

I'm afraid your comparison is invalid. While I'm sure you're right for your own environment, the one we're talking about here is a company hosting 19,400 web sites of which the majority are based on dynamic content. I don't know how many machines they have, but at their scale, they'll be employing a reasonably large number of full-time admins. The fact that existing employees might have transferable Windows skills is of no relevance at all, because they'll have employees dedicated to the task.

Sadly, I'd have to disagree with this troll

[mrjb]

All you need is Windows experience (which everyone has nowadays). Linux requires special knowledge and/or training. Why it is that on Windows you call it experience, whereas on Linux you call it 'special knowledge'? It is the same thing.

Linux isn't user friendly enough for the average workplace drone to administer. ...and nor is Windows, if you want to make sure it's done right. It requires some "special knowledge" to get things right.

most small/medium companies can't afford to hire somebody extra purely because they have Linux experience.
As you mentioned, you don't need 'someone extra'- just employees that are more versatile. This will in fact help you to operate with less people. Replace the first two techies that leave your company with people that (also) know Linux and you're set.

Of course, if your company does have people familiar with Linux, then the TCO is going to be WAAAAY lower.
Q.E.D.

Cost to consumers is lower on linux

[Respawner]

The company in question according to the article is hostingbasket, so lets check it out: Linux and Windows.
Their "basic" packet starts at 8 euro / month for linux hosting and 10 euro/month for windows hosting
Now, how about hostingbasket and microsoft ? well, I'm going to be honest, I don't know much about how msn works but this company has its own subdomain on msn http://hostbasket.msn.be/ , so this study looks a bit odd to me

If Windows really is cheaper to run

[petard]

Then hosting companies are behaving irrationally, including the one that is the subject of the study. If it costs less to operate a Windows platform than a Linux platform, then the hosting companies would rather have their customers use Windows platforms. This would lead them to price the Windows packages lower than the Linux packages; after all, if windows were cheaper, they could do that and still be more profitable.

Hostbasket, the subject of this study, is not doing that:
Their basic Linux package is 8 Euros/month and their basic Windows package is 10 Euros/month. So if the results of the study were true, this pricing scheme would be quite irrational.

Looking at other hosts, this seems to hold up. interhost wants 19 GBP/month for Linux versus 25 GBP/month for Windows. Over at New York Internet accepting all the defaults for their BSD plan nets a quote of about $42/month; a similar Windows-based plan is $64/month. And over at hosting.com their managed hosting plans for Windows servers start at $230/month while the same plans for Linux start at $195/month. I was able to find, over at 1and1, shared Linux hosting and shared Windows hosting that cost the same.

I was not able to find any provider that offered cheaper Windows hosting than Linux hosting.

So, assuming that everyone behaves rationally, if the numbers in this study were accurate at all, the hosting provider that is the subject of the study would offer cheaper Windows hosting than Linux hosting. They don't. If the numbers in this study were generally applicable, you'd find that most hosting providers who offer both would offer cheaper Windows hosting than *Nix hosting. They don't. I can only conclude that the study is bogus in some way and shouldn't be trusted, since it fails to predict rational behavior in a very open marketplace (i.e. one with very low barriers to entry). Businesses are very good at thinking with their wallets, and if this study were true then there's a huge money making opportunity that everyone is letting go.

Sure...

[Dion]

... if you're a digger operator.

Certifications are worthless.

What matters is real-world experience and before you have any of that there is formal education to get a foot in the door.

In the company i work for there is noone who has any certifications, execpt for a few people who got them by accident before being hired, there are plenty of good people who know everything there is to know about Solaris, Oracle, Linux, Java and C++ though.

What is a certificate good for other than to show that you can think inside of that particular vendors box?

Hmmm... ? Huh? What?

[Fallen Kell]

Well, it takes me approx 1.5 hours to setup a new webserver from scratch (i.e. no OS installed, no formated or partitioned disks, etc., nothing), which runs Solaris 10 (I know, it is not OSS yet, but will be soon), my own custom compiled Apache from latest source release, PHP, and MySql.

Trying to figure out how that is more expensive then Windows. If anything, I just saved myself 6 hours of patching the OS from WindowsUpdate (update, reboot, update for the updates, reboot, update for the updates to the updates, reboot, update one more time, reboot, check that there are finally no more updates). That is correct everyone. Last time I installed Windows XP Pro, it took 4 windows update sessions before there were no more patches left, and it was an SP2 install disk as well, just imagine if you had the original WinXP Pro disk, add 2 more windows update rounds to that number (SP1 and then SP2).

It's not the 1st server, it's the Nth server

[ElectricRook]

It's not the first server that is cheaper, it's the N'th server that's cheaper. To drive a mouse around on one MicroSoft server is easy, little training required. The second is only a little more expensive.

The real economy is when you have big numbers of servers. When using cron to drive your admin scripting and SCP to change that scripting. That's when UNIX varieties really shine.

An army of admins are needed to admin a bank of Microsoft servers, one or two smart admins can easily handle several banks of UNIX servers, and have time to contribute to /. too.

especially at upgrade time.

Been Both Places

[W. Justice Black]

I've been a professional Windows SA and a professional Linux SA in various parts of my career. Relevant observations follow.

Installation:

Windows - Nearly trivial if all you care about is MS tech and don't need a database. Somewhat less so if you need, say, php and a database. Integration can be mitigated across several systems via Ghosting. Er, not really server-side. Ghosting IIRC is rather verboten in Microsoft's mind.

Linux - Trivial if you use the distro's packages. Significantly less so if you need to integrate, say, Tomcat with Sun JVM or Oracle. Integration and configuration can me mitigated across several systems via configuration management (cvs, svn) or via scripting or via just copying working configuration files to server #n+1.

Configuration:

Windows - Simple if you're not doing anything terribly interesting (and most people don't). Configuration replication is significantly more difficult. Incremental configuration changes (e.g. adding another site) can be scripted if you REALLY know what you're doing or are using third-party tools like Plesk.

Linux - Somewhat complicated if starting from scratch, especially with Apache 1.3 and single config files. Easier if starting with Apache2 and separate config files. Integration of third-party things can be somewhat difficult. Easy to "roll back" changes using a configuration management system, and relatively easy to script incremental configuration changes.

Updates:

Windows - Easy for base system via Windows Update. Somewhat more work for third-party components.

Linux - Easy for base system and perhaps all components that would be considered third-party above. Somewhat more work for third-party components (but the list of "third party components" is smaller than that for Windows, as PHP/MySQL/Postgres/Whatever are part of the distro).

Performance:

I think the endless performance arguments are counterproductive. Linux "feels" faster, but that's not quantifiable, and there are countless ways that tests can be structured to optimize for one architecture or another, especially once you toss application layers (xxMP, Tomcat, CF, etc etc) in there. If performance really matters that much, an organization probably has enough resources that they can make a better evaluation for their payloads than politically-motivated third parties anyway.

Conclusion:

I'm not really going to say anything that others haven't said better elsewhere. If you're looking at one departmental or small business web server, Windows is probably easier to start out with, especially if you don't have the talent to grok Linux right off the bat (that gap is shrinking year-over-year, but it's still there). Once you're looking at any real scale (and want to do things like actually replicate configurations and the like), Linux is far more useful and probably cheaper in scale.

That said, Hostbasket itself charges less for its Linux offering than it's Windows one, and (at the most conservative), Windows is more expensive in every area except labor and (bizarrely) bandwidth if you multiply out the percentages with the calculated TCO number. They're showing Linux as 3.5x more expensive in labor than .NET 2.0, which is dubious in my mind. There's a story here that's not being told--3.5x is a huge jump and there's got to be either a juicy story here that looks bad for Linux (unlikely, or it would've been publicized) or something structural that may invalidate the whole study (more likely, by elimination).