Apache Web Server Share Falls Below 50 Percent For First Time Since 2009

darthcamaro writes "Apache has always dominated the web server landscape. But in August, its share has slipped below 50 percent for the first time in years. The winner isn't nginx either — it's Microsoft IIS that has picked up share. But don't worry, this isn't likely a repeat of the Netscape/IE battle of the late 90's, Apache is here to stay (right?)" The dip is mostly the result of GoDaddy switching to IIS from Apache. Which is to say GoDaddy hosts a whole lot of sites.

Comments

note to self..

[Anonymous Coward]

..another reason not to host on godaddy.

Re:note to self..

[CFD339]

You needed one other than the hostile upsell pressure virtually every where they touch?

GoDaddy IIS

[naubrey]

Which is to say that GoDaddy hosts a lot of *parked* domains on IIS.

Re:GoDaddy IIS

[Manfre]

Which is to say that GoDaddy hosts a lot of *parked* domains on IIS.

...which were previously served using Apache. None of these stats will ever be able to convey the usefulness of site content based upon web server software.

Re:GoDaddy IIS

[ShanghaiBill]

Which is to say that GoDaddy hosts a lot of *parked* domains on IIS.

Honest question: Why did they switch? I have never understood why anyone would use IIS, and always assumed ISS users were clueless newbies. So why would GoDaddy go to the time and expense of switching? What do they gain?

Re:GoDaddy IIS

[Trepidity]

You could look at what the share is among the top N domains, for N=1000 or N=10,000 or whatever, at least as a sanity check.

Re:GoDaddy IIS

[whoever57]

Honest question: Why did they switch?

My WAG is that MS threw a bunch of money at Godaddy, not directly, you understand, but indirectly.

Furthermore, my conjecture is that MS is prepared to throw this money at Godaddy because Microsoft's share of sites was looking rather sad (3rd place for market share of active sites last month).

Re:GoDaddy IIS

[csumpi]

Simple: asp.net. Plenty of half assed coders out there can, with little effort, build a website using Visual Basic or C#.

Sounds like a great accomplishment and major win for Microsoft.

Re:GoDaddy IIS

[Pseudonym]

Your half-assed crap code. Our passion.

Re:GoDaddy IIS

[Sir_Sri]

That's not necessarily a good metric either, as systems built for that amount of traffic are not necessarily indicative of what is suitable for the rest of us. That's sort of the formula 1 versus a regular driving vehicle problem.

Re:GoDaddy IIS

[KingMotley]

Simple reason really. Microsoft serves static pages faster than Apache and scales better under this scenario. It allows Go Daddy to park more sites on the same host, which then saves them money.

Re:GoDaddy IIS

[Bert64]

If you look at the netcraft graph going back several years, you will often see significant bumps in share either for or against IIS. Several of these are down to MS paying large hosting providers to put their parked sites on IIS for promotional reasons.

Re:GoDaddy IIS

[xyra132]

It might be a good indication of what the mainstream hosts will be using in a couple of years though. As with your formula one analogy, (a subset of the) technology developed at the extreme high end is commoditised and trickles down.

Re:GoDaddy IIS

[Ash-Fox]

Unfortunately, I have no graphs I can share with you due to confidentiality reasons. You can get Apache to outperform Nginx by configuring Apache to use prefork, then increasing Linux's file descriptors to a much larger setting than default to coincide with the forked processes and the connection limits you wish to handle.

This is something that is done in large cloud setups used in major websites that use Apache with load balancers, or in more sophisticated networks, using BGP to perform "true load balancing" (requires a router to hash source and destination IP to maintain a connection to the same server and weight calculations).

I have not managed to get Nginx to perform as well as Apache in this configuration (prefork + increased file descriptors), or any other configuration when Apache is in this configuration. IIS in my experience doesn't do too badly with static content, however, I have never seen it outperform Apache in the 'proper' environment and configuration when it came to static content (this also included disabling modules not in use), however the differences were very minor.

I will say though, if you're using IIS for serving static content, you might not be using your money wisely. It's not really as cost effective to pay for windows licenses and machines verses just the machines (I expect GoDaddy got freebies from Microsoft). Before the argument on support - You will often find that hardware vendors like HP and IBM provide their own support for various operating system configurations and certain uses on server hardware is included with your hardware contracts, should you be a sufficiently large buyer.

It really only makes sense to run IIS if you have specific IIS/windows applications to run on the webserver like .NET applications. Costs could be reduced in such a scenario to keep IIS for static pages if it required a new breed of administrators to join a team to maintain some 'free' server option. However, if the administrators are multi-platform (like where I work), this is unlikely the case.

Re:GoDaddy IIS

[benjymouse]

Unfortunately, I have no graphs I can share with you due to confidentiality reasons. You can get Apache to outperform Nginx by configuring Apache to use prefork, then increasing Linux's file descriptors to a much larger setting than default to coincide with the forked processes and the connection limits you wish to handle.

Using processes to serve individual requests seems like a massive waste of resources, and I have a hard time believing this will run faster than a standard threaded worker. Do you have any theory behind why Apache would process requests faster with preform as opposed to worker?

I have searched for optimization guidelines for Apache and I cannot find anyone recommending this setup, except for when compatibility is required.

I suppose that if Apache is used to serve non-static content using a foreign/3rd party module, there could be local lock contention issues if the module is poorly written. If the lock contention is intra-process (process local locks) I suppose that you could get around that by running multiple processes. But that would merely mean that the server is not being slowed down waiting for locks, not that it will run faster than Nginx.

But in this case no such modules are involved, and lock contentions are clearly not at issue for any browser in the bandwidth saturation tests.

Hmm

[Tailhook]

The statistical effect of millions of empty, neglected GoDaddy hosted sites will not ultimately mean a great deal. It does raise a question for me, however; what benefit does GoDaddy hope to realize with IIS? My last contact with IIS was about 9 years ago. At that time it was fragile, insecure and plagued with mysterious "metabase" corruption problems. The thought of using such a thing for large scale hosting seems absurd and I've ignored it ever since.

Has it since improved enough to entice really large operations?

Re:Hmm

[ackthpt]

No doubt it has improved, but it's still a PITA to work with. I have to work with it now and really miss Apache.

Re:Hmm

[loufoque]

IIS runs on Microsoft Windows.
GoDaddy administrators do not have the skill to manage Linux boxes.

Re:Hmm

[readingaccount]

You basically just admitted that Linux boxes are harder to administer than Windows servers. This makes Linux servers much less appealing for companies when you can find Windows server admins for a dime a dozen, but Linux admins are harder to find and generally cost a lot more.

Re:Hmm

[DuckDodgers]

Linux isn't harder to administer because of any inherent problems, it's harder for the average person to administer because we probably had Windows in the home computer and at the office, and at the school. Making the jump from Windows 95 to Windows NT or from Windows Vista to Server 2008 is a lot easier than jumping from Windows to Linux.

If you're a serious power-user administrator, Linux and Unixes in general has been easier to administer than Windows Server for a very long time. You have more interoperable shell tools at your disposal. The Server GUI is better for an admin novice, but terminal tools are quicker for a power user than toggling through programs and hunting through menus. Microsoft is catching up with PowerShell, but even if the technology is extremely flexible and mature (and it may well be), they took the odd step of inventing a new syntax different enough to be confusing to people comfortable with bash or cmd.exe - me among them. Now I'm asking myself whether making the investment in Powershell is worthwhile. It probably is, but I don't look forward to it.

Re:Hmm

[jader3rd]

Now I'm asking myself whether making the investment in Powershell is worthwhile.

It's worthwhile.

Re:Hmm

[evilviper]

Windows servers undoubtedly have the advantage of being able to turn up a service almost on accident, and have it minimally work. Actual administration and maintenance of them, though, is a Kafka-esque nightmare. I feel bad for Exhange admins. I've heard many horror stories of Windows support telling Admins there's no fix, no fallback, and they'll have to reinstall the entire server recreate datastores, and then they take a few months manually importing All user emails.

Linux/Unix servers take more knowledge and effort to get up and running in the first place, but then are much more stable and deterministic, handle higher load, need less babysitting, and are easier and more consistent to keep updated and make changes to, knowing you're never going to have unrelated services break, or mysterious slowdowns and service unavailability.

There's no doubt what comes out ahead in the end... Linux adminsa can mantain many times more servers than Windows admins. Consider that those Windows admins won't be free, and you'll be cash positive by hiring Linux admins in a very short time. I've worked for some of the most penny-pinching tight-wad companies around, and they emphasize Linux heavily (including on the desktops) paying their Linux admins more than even most management, and yet they heavily prefer Linuxx, and wouldn't dream ofusing Windows for anything important.

Re:Hmm

[smash]

+1. Powershell is quirky. It is prettty excruciatingly slow. There are bugs. But it really is pretty neat, and nothing similar exists in the Unix world as yet. If you need to admin Windows boxes, you'd definitely be well advised to learn powershell.

Re:Hmm

[smash]

Not really true. I've seen Linux boxes administered by muppets, and back in the day when I was 20, I was one of those muppets in charge of looking after a fleet of Linux boxes. Just because someone can use google and click their way through a redhat installer, doesn't mean they have a clue.

Platform choice is pretty irrelevant as far as judging competency goes actually, IMHO being a competent administrator/architect is more about change management (i.e., how do we get from A to B without fucking everyone over), engineering resiliency into your designs, being proactive about security and using the best tool for the job, in a platform agnostic manner.

Trade-offs will be involved (this is essentially what engineering is). If (for example) running a Windows web server makes it 10x easier for your internal web development guys, and it can be secured by spending a little more time than an apache box, then you run a Windows web server, and stick whatever content-aware firewall you deem appropriate in front of it.

In other situations (e.g., DNS servers, firewalls, mail relays, etc) - Linux, BSD or other unix platform of choice may be more appropriate.

Re:Hmm

[DaveAtFraud]

You basically just admitted that Linux boxes are harder to administer than Windows servers. This makes Linux servers much less appealing for companies when you can find Windows server admins for a dime a dozen, but Linux admins are harder to find and generally cost a lot more.

Er, no. Windows makes the easy things easy (pick what you want from the list rather than, horror of horrors, type something) but still hasn't succeeded in making the difficult easy. This lulls people who think they know what they're doing into jumping into the deep end and finding out they can't swim. Lots of things when setting up a server (web or otherwise) that require an understanding of the underlying networking. The Windows admins who don't know this are the ones who are "a dime a dozen." The ones that do can create a secure, functional site with Windows but wish they had Linux since it's easier and more secure and faster and more flexible and....

Cheers,
Dave

Re:Hmm

[MightyMartian]

My experience with Powershell is sufficient to state that Windows users can keep it. Bash is a far far more mature shell with a helluva more lineage and experience behind it.

Re:Hmm

[siride]

Yep. And if you want OO or real programming, just use Perl or Python.

And for God's sake, the whole signing scripts business with Powershell is a tragedy. I can understand the value of being a little bit more tight with scripts that can do harmful things, but it should only matter for scripts that need to run as admin or do system management tasks. I shouldn't need to cryptographically sign a script to extract tags from music files, for example. The process to do the signing is itself unnecessarily complex.

Re:Hmm

[smash]

I'll grant you bash is more mature and intuitive, sure - but you can do things in powershell that just either aren't possible with bash without writing helper applications in a non-scripting language or are exceedingly convoluted.

The big difference is the object pipeline which takes a little to get your head around, but enables you to do far more processing on data than text manipulation with sed, grep, awk and friends.

I'd suggest opening your mind a little and giving it a shot. If you don't administer windows boxes, fair enough its pretty pointless, as the whole point is interfacing to .net objects which don't exist on Linux.

But if you DO need to deal with Windows boxes, you'd be doing yourself a dis-service to write Powershell off.

Re:Hmm

[turbidostato]

"Technology changes a lot in 9 years"

Not 9 but 20 years ago I run NFS and CIFS, LDAP, Bind, Postfix... now I run NFS and CIFS, LDAP, Bind, Postfix...

No, technology doesn't change a lot, marketroid guys make it look like so to stay in the business of selling new licenses.

Re:Hmm

[benjymouse]

And for God's sake, the whole signing scripts business with Powershell is a tragedy. I can understand the value of being a little bit more tight with scripts that can do harmful things, but it should only matter for scripts that need to run as admin or do system management tasks. I shouldn't need to cryptographically sign a script to extract tags from music files, for example. The process to do the signing is itself unnecessarily complex.

Just set the execution policy for the scope that you want. Type man set-executionpolicy -para scope. You will notice that scope can be set for the process, the current user account or for the local machine. So if you want to then simple set a less restrictive execution policy (like RemoteSigned) for your current user. That will still prevent scripts downloaded through a browser or received through a mail to be executed.

There are a lot of legitimate uses for script signing. For instance, for a tightly managed system you could put in place a regime where an auditor/reviewer must sign a script only after review, and the farm of machines will only execute scripts that have been signed by the reviewer. Thus you can have devops develop scripts and test at their test systems, but to deploy the scripts they must follow the review process.

Re:Hmm

[benjymouse]

There are *a lot* of little things in PowerShell that makes you go "aw, that's a good idea". Things you will not find in other shells and neither in Pytho, Ruby or Perl.

Off the top of my head:

* Consistent common parameters for "impact management": You can pass a parameter called -WhatIf to every built-in command that may change persistent state. The -WhatIf parameter runs the command in simulated mode, only echoing on the console what it *would have* done. Similarly a common -Confirm parameter which asks *before* changing persistent state. It even works for scripts and functions: If you declare that your script (in a .ps1 file) "supports shouldprocess" you can pass the -WhatIf parameter to your script and PowerShell will set the whatif preference for the duration of the entire script - as if each command of the script had been passed a -whatif parameter as well.

* Commands, functions, script blocks and script files declare parameters with (optional) static types. This information is used by PowerShell to coerce values to the correct types before invocation. But the declarations can also contain declarative validation attributes, allowing the *shell* to validate parameters before invocation. Declarative validation can validate required parameters, string lengths, number/date ranges, regular expressions, value sets. The kicker here is that the script author does not need to *implement* validations, merely declare them, the information is available to the shell which can use it to both validate parameters before invocation, but *also* to report the validations through the help system. That's right, when you set up validation, help text that describes the acceptable values is automatically generated from this meta information.

And yes, even the tab completion (or intellisense in the integrated scripting environment) will pick up on the parameter type and validation. If you restrict a parameter to a certain value set, tab completion will cycle through those values when the shell determines that you request tab completion for the parameter.

* The PowerShell help system allow for in-script help text through special code comments. No need to author external help files. You can write the documentation right there in the script (using special "dot" comments), and when you do man myscript.ps1, the help system will report the documentation.

* PowerShell workflows allow scripts to suspend and resume at a later time. No, this is not the process suspend of sh shells. PowerShell actually saves the state of script execution to persistent storage and you can resume execution later, even after system restart - or on another computer. This is incredibly useful for the type of scripts that manages farms of servers and that may be running for a long time. If the script is somehow interrupted (power failure, hardware failure) it can later automatically pick up its execution from the latest savepoint. I.e. you can restart it and have it run to completion.

PowerShell is not simply a programming language. It has many features which are directly aimed at being used in a scripting setting and which are not found in general purpose language like Python or Ruby.

Re:Hmm

[MightyMartian]

I'm going to be blunt. I have yet to meet a text replacement/massaging problem that couldn't be solved with ask. Years ago I wrote an ask script to translate a gawdawful mainframe export of a stationary supplier's catalog. W're talking tens of thousands of records, all space delimited with variable field sizes for different kinds of inventory records and some records that were even multi line.

I wrote the script on my Linux machine at home, grabbed the script and a compile of gawk for DOS and took it to the customer, wrote a quick batch wrapper and we took the whole bloody file and imported it into his POS system.

I'm sure powershell is a wonder, but those who make grand declarations about the limited utility of *nix userland utilities are either ignorant or for dubious reasons sweeping away decades of *nix scripting.
I'm sure

Re:Hmm

[smash]

Dude... Powershell is already coming up 7 years old, and the core concepts are still the same, there are just more cmdlets available. If you're even comparing to Ruby, Bash or Python you clearly have no idea how powershell operates or what it can do and have not spent any real time playing with it.

And as you have not spent any time playing with it, your opinion is entirely uninformed, and you're just shitting on it because it is written by Microsoft.

Maintaining the status quo because this is how we've always done things is retarded.

So, check it out or don't. Not really concerned. But yourself, and the open source world in general would be well advised to check out anything new by any commercial developer, including microsoft - and profit from the mistakes and successes made without reinventing the wheel yourself and making the same mistakes.

Re:Hmm

[DuckDodgers]

Thanks for filling in some details. I've heard people rave about PowerShell before. I've also read that Microsoft planned adding an optional view pane to every control panel and administrator tool in Server 2012 that would output the PowerShell equivalent command to anything you did in the GUI. That strikes me as a brilliant way to turn your slow click-monkeys into fast shell admins. I'd like to see Red Hat or Canonical (Ubuntu) do something similar for Unix, but I don't think even Red Hat has the engineering resources for a project like that.

The problem I had with powershell, which I assume is commonplace, is that I fired it up expecting it to be a backwards-compatible superset of cmd.exe. It isn't, lots of the syntax that works fine in cmd.exe gives errors in Powershell or works but does something different from the cmd.exe equivalent. That of course makes me nervous - to my knowledge, bash hasn't changed much since introduction over 20 years ago. It has arcane syntax and plenty of warts, but an investment in bash will probably still be useful in another twenty years. Will Microsoft be using Powershell in ten years, in a form that's compatible with current syntax?

And while consistency in command interfaces is hugely powerful, there's something to be said for the global interoperability of using text everywhere. If I'm doing complicated administrative tasks on Linux, I can do part of it in C++, pipe the output through a sed script, pipe that output through some Python script I downloaded, pipe that output through a Perl script I wrote, and feed that into some Ruby program. Then I can take the Ruby results and put them through a Basic application and then through Common Lisp and then a C# program running on Mono. Of course that's a contrived and very silly example, but the point is that while text is far more tedious to work with than (properly designed) object types, you have easy interoperability between hundreds of different tools.

Re:Hmm

[smash]

Powershell can invoke any .net object's methods. Given that powershell is only really available on Windows, which is where you'll be using it - it is free on that platform, so bash being free doesn't really make any difference if it is a windows box (or network, given that almost all cmdlets can be used on remote machines with the trivial use of -computername parameter) you're administering.

Re:Hmm

[DuckDodgers]

I think you unfairly got modded down, sorry enthusiasm for FOSS got in the way of fair discussion.

Typically Linux gets more patches because the person has a lot of software installed. If you've got a pretty barebones setup and only pulled in exactly what you need (Apache + PHP + PostgreSQL), I imagine the patch rate isn't too different from something like IIS + PHP + PostgreSQL or IIS + .NET framework + SQL Server.

Re:1st post.

[ackthpt]

apache 4 life!

No kidding. I hate IIS right now. It's so much more time consuming to sort out configuration issues with than Apache.

Removed parked sites

[Todd Knarr]

In my book, the stats ought to be excluding "parked" sites, ones which don't have any content beyond a parking page. I'd also exclude sites whose only content is boilerplate advertising (eg. the one you get if you're on Cox Cable's internet service and type a nonexistent domain into your browser). I'm more interested in what servers are being used for productive work without the numbers being skewed by the guy who registered 10,000 domains related to the latest fad and is waiting to see which ones he can sell at a profit.

Re:1st post.

[Anonymous Coward]

But IIS is NSA-friendly!

Re:1st post.

[Anonymous Coward]

There hasn't been any serious security holes in IIS for years now. So the government ordered MS to add PHP support.

Not the first time either

[MrNemesis]

Dupe! ...and the knock-on.

I'm beginning to wonder if GoDaddy's web server policy follows the solar cycle... :)

From the look of Netcraft's graph, prior to the GoDaddy move it looked like most of the marketshare lost from apache went straight into nginx (itself also frequently used as a caching proxy/frontend to another web server on the backend) so I'm not quite sure what the summary/TFA are trying to imply.

http://news.netcraft.com/archives/2013/04/02/april-2013-web-server-survey.html

Rubbish

[Tough Love]

it's Microsoft IIS that has picked up share.

No. Microsoft picked up a bunch of parked domains and its long term trend is still down, even for parked domains. In terms of active sites, Microsoft's trend is steadily down, now around 12% and sinking. And it is indeed nginx that is mainly picking up share from Apache, though Google is hanging in there pretty well too. This puff piece glosses over the one fact that can't be denied: Linux servers rule the web by a large and increasing margin.

http://news.netcraft.com/archives/2013/08/09/august-2013-web-server-survey.html#more-12060

Re:base it on traffic vs. how many domains host'd

[93 Escort Wagon]

Netcraft's report shows the percentages for all domains as well as for active domains.

This article is a bit sensationalistic - no surprise. As a percentage of all domains, Microsoft is at 23% (Apache's at 47%). Looking at just non-parked domains, they're at 12% (versus 54% for Apache). Not really much of a "Apache vs. IIS" story there...

If there's any news at all, it's that servers other than Apache and IIS have managed to gain significant traction over the past couple of years. I remember when it had really turned into a two horse race, and gains by one exactly mirrored losses by the other. But now it's a bit more of a healthy competition.

Re:Citation needed - When/why did GoDaddy switch?

[dltaylor]

The originally-linked Yahoo finance page is expired, but here's the /. discussion from 2006:

http://slashdot.org/story/06/03/23/008229/godaddycom-dumps-linux-for-microsoft

Microsoft was paying large hosts to switch $10 sit

[raymorris]

A while back Microsoft was paying hosts and registrars with large numbers of domains parked, or $30 / year type, to switch over.
I don't know if that program is still active.

Uh, didn't godaddy switched over years ago?

[the_B0fh]

why is it that everytime I read about a dip in apache stats, it's because of godaddy switching over? Bloody hell, they've been switching over for years, just how many effing sites do they have?

Re:1st post.

[smash]

Be that as it may (I hate the IIS administration interface as well), for an enterprise who runs microsoft on the desktop, microsoft SQL, and other microsoft services, IIS integrates far easier into that environment.

And I suspect this is where it is winning share - the web isn't static pages any more.

Sure, Apache can do this, but the environment is totally foreign to your average corporate type.

And as usual, security is probably some way down the priority list.

"admitted" 3rd graders can reboot Windows $4 hosti

[raymorris]

No, he "admitted" than any 3rd grader can reboot Windows. $4 hosting companies don't get server admins, the get phone monkeys. I used to get frustrated with their "admins" being clueless, but then it happened. I was working with HostGator, a top hosts who has the same business model as GoDaddy hosting, and I found out their "admins" don't have access to the datacenter. They are literally just a phone bank and marketing company, with The Planet running the servers. So yeah, it's easier to hire Windows phone monkeys than Linux phone monkeys. (Maybe because Linux users tend not to be the phone monkey type?)

If you want actual qualified admins, people who know the difference between a gigabit and a gigabyte, you're going to pay no matter which OS. (Though I do know a _certified_ Windows admin who doesn't know the difference between bits and bytes ...)

Re:1st post.

[MightyMartian]

IIS is an absolute fucking nightmare when you have to deal with a buggered up config. Actually that applies to most MS point and click services. Apache can be a bastard, but at least I can back up the configs with a quick "cp".

Worst experience I ever had was with IIS and Exchange and something going wonky with IIS's settings, and OMA completely screwing up. In the end I literally had to uninstall IIS. Only MS would build things with such fragility and such insanely dangerous solutions.

fork() vs epoll()

[NynexNinja]

I think when Nginx first came on the scene (a little bit after libevent was released), Apache had known about the scalability problems associated with using fork() versus epoll(). This was almost a decade ago. Apache has yet to provide a scalable implementation using epoll similar to what Nginx provides. Its at least a 10x speed improvement on the same hardware.

All that I can say is that all new installations over the past I'd say about 5 years, I've been doing using Nginx only because Apache just can't scale well with their fork() implementation compared to Nginx. I'd say this has something to do with people leaving Apache, at least all the people I know.

Re:fork() vs epoll()

[codealot]

Why are you still using prefork? You have at least two good alternative MPMs, one of which can use epoll().

it is getting a LOT better

[raymorris]

Having acknowledged the sometimes extreme security issues PHP has had in the past, I have to say it's getting a LOT better. PHP was designed as something like a blogging system, not a general purpose programming language. Because people are using it for general programming, they have made huge improvements.

Now if only people would read the giant warning at the top of the SuExec documentation: "SuExec can result in severe security risks. Do not consider using SuExec unless you are knowledgeable about ...". That warning is there for a reason. SuExec / suPHP really is dangerous as hell, just like it's documentation says.

Re:it is getting a LOT better

[Anonymous Coward]

The problem isn't the language, it's the community. Google "php mysql" and the first link teaches you how to create a SQL Injection point. And 'experienced' PHP developers still write code that way. I'm convinced they just don't care.

Re:1st post.

[Anonymous Coward]

Sounds like your gripe should be with the fine PHP people, not the Apache project.

Actually, they are losing share to nginx

[haggholm]

The Netcraft article does have statistics that exclude parked domains, and here IIS doesn't look to have an increasing trend at all. The only webserver with a steadily increasing trend is nginx. In the graph of the top million busiest sites, nginx is again growing the fastest, though "other" is also a growing category.

Re:1st post.

[micheas]

It was mostly due to microsoft cutting a check to godaddy to not show apache traffic server in the headers.

Godaddy runs IIS on linux. Well, they run IIS behind apache traffic server so which webserver to count as the webserver is a bit of an academic question. The moral here is that godaddy hosts a lot (hundreds of thousands, if not millions) of inactive sites that they collect 9.95 or so a year for hosting.

Re:1st post.

[Zenin]

You can backup IIS's config just the same. It's just an XML file (and a surprisingly easy to read/understand one at that).

You can also do your config by editing it as well, although typically you'll use something like AppCmd or more modernly PowerShell.

It's frankly easier to reliably automate/script IIS configuration changes than Apache. Apache's configuration system is incredibly powerful and at times that's needed, but that power also means it's effectively impossible for a random admin script to make sense of it enough to modify. Such a tool must intrinsically know not just Apache's config system...but your specific implementation with it. AppCmd and PowerShell can pretty reliably walk into nearly any IIS setup, no matter how convoluted, and safely make additions, tweaks, etc.

Frankly I'm first and foremost an Apache fan, have been since it was literally A Patchy Server. And I still deploy it more often than not, often in front of IIS to get some clever hack done that just isn't practical in IIS.

But that said...I'm warming up to IIS, especially as C#/.Net gains major traction in the wake of Oracle's kiss of death to Java.

Re:base it on traffic vs. how many domains host'd

[93 Escort Wagon]

Actually, looking at the raw numbers... it's just shy of 90%! But even for Apache, something like 70% of sites are not "active" by Netcraft's metrics - and it's a similar story with all the others.

Re:Microsoft was paying large hosts to switch $10

[gbjbaanb]

IIRC, GoDaddy switched to IIS for these parked domains and a dip in Apache usage appeared, then reversed itself a year or so later... now its repeating.

Seems more like a money-making initiative fromGoDaddy, or a money-losing initiative from MS yet again. What's the chances history will repeat itself once the contract runs out...

Re:1st post.

[TheRaven64]

And people say Microsoft doesn't innovate. Making something that's more painful to configure than Apache requires an impressive amount of R&D...

Re:base it on traffic vs. how many domains host'd

[TheRaven64]

And on active sites it looks like IIS has been dropping in use since April 2009. Apache has actually picked up a bit since they had a big drop between 2005 and 2007, although it's down a bit since the small peak in 2011. The most interesting part is the growth of the 'other' line from about 6% to about 12% over the past 3 years. It's sad to see Lighttpd has almost died (not much development for years, now just counted as part of 'other'), but it's good to see a few different servers clustered around the 5-10% mark. Having a single server with over 50% of the market makes it quite an attractive target. It would be good to see more diversity.

Prefork is the worse MPM for performance

[kervin]

Prefork plus increased file descriptors? You're kidding right? While you can get Apache to match NGinx, it's definitely nowhere that simple. As optimized as Unix fork() is, processes are going to use more resources than threads in this scenario every time. Prefork is the worse MPM you can use when you need performance. Even the Apache manual spells this out.

You'd have *begin* with worker or event MPM, use Apache 2.4 at least, and finely tune for your Application and specific load.

The benefit of NGinx is that you get a highly optimized web server right out of the box. You don't have to mess with the configs and you're almost there.

Technically the Apache team can do the same if they get rid of Prefork and a whole bunch of decades old legacy configuration options. Remove code processing modules from the webserver application space, i.e. get rid of mod_php for php_fpm, etc. All this can be configured now and you'll get that speed and stability, but it's just not done out of the box.

With NGinx it is. The only way to do things is the 'fast' or optimized way.

Re: Prefork is the worse MPM for performance

[Ash-Fox]

processes are going to use more resources than threads in this scenario every time.

I definitely agree it uses more resources typically - It doesn't however mean that the response time is worse.

Prefork is the worse MPM you can use when you need performance.

System resource wise, not necessarily performance.

The benefit of NGinx is that you get a highly optimized web server right out of the box. You don't have to mess with the configs and you're almost there.

Flipping a few configuration settings in Apache's httpd.conf and Linux's sysctl.conf is trivial for me, it's only a minute or two.

Remove code processing modules from the webserver application space, i.e. get rid of mod_php for php_fpm, etc. All this can be configured now and you'll get that speed and stability, but it's just not done out of the box.

mod_php, php_fpm aren't part of the default Apache configuration? And if you want to remove module support all together, you could just compile Apache with:

./configure --disable-so

Not that I have ever seen any notable speed differences through disabling Dynamic Shared Object (DSO) Support.

With NGinx it is. The only way to do things is the 'fast' or optimized way.

Load testing performed through SOASTA Cloud Test sadly didn't live up to those numbers on my environments. There was no notable improvement over Apache and at the cost of sacrificing useful functionality provided in Apache for what appeared to be no gain. Perhaps it was my environments, but I doubt it.

Re:base it on traffic vs. how many domains host'd

[IwantToKeepAnon]

The Netcraft graphs show "Google" as a platform, last I read (a couple years ago granted) that google ran a customized version of apache. Does google now offer a honest-to-goodness webserver of its own? Or can googles #'s be added into the apache category? And what about Tomcat sites? Are they already included in the apache #'s?

Thanks for the link.