Hacker News Reader

I'm going to throw a possibly unpopular view on the table here: sticking all this on Linux is purely about cutting costs but saves nothing.

The administrative overhead of deploying all this stuff is incredible as are the disparate infrastructure and software models. You might be saving 11x on an Azure instance, but your ops time and staffing requirements will cost more than that. Throwing something on Azure is really cheap with respect to time and effort. If you need to scale past that or cost becomes a limitation, you might as well fish out for a Rackspace or other dedicated package.

We're at the £50k hosting budget a month mark now (1x production DC, 1x hot spare DC, 2x packed 42U racks in each) and 2 meat based products to manage it all. Not much of that cost is software licenses.

Note: I have nothing against servicestack - in fact it's rather good (we use servicestack.text)

For ref, I turned down a position at 7digital a couple of years ago as I though they were smoking crack when stuff like this was mentioned.

Unless you're new to the web, you'll be surprised to learn that most of the Internet runs on Linux. It runs because it's a commodity server, because it has first-class support for most NoSQL data stores, caches, proxies, etc. because its highly configurable, because it's much more automatable where everything is scriptable with simple unix commands.

Going to call BS on Azure being cheaper. All of ServiceStack's live demos are deployed using simple bash scripts. There's none of these admin overhead costs, I check-in on my dev box and run a remote script to deploy straight from the Git repo. The projects are also un-touched ASP.NET or stand-alone Console hosts (which I host behind an nginx reverse proxy). I've no doubt that it would be at least 11x more expensive if I decided to host servicestack.net on Azure, even more if I decided to use SQL Server (PostgreSQL/MySQL is free to use on Linux). And I'll also miss out on cool linux utils like rsync which is how we migrated GB's of content onto the new vServer.

The indexable "world wide web" of startups, big tech brands, VC, video sites, porn and blogs about cats falling out of windows, yes.

When it comes to the deep web i.e. corporate applications, business tools etc, it's just not seen at all. These are the things that produce a lot of value and solve a lot of problems. I'd argue they are cumulatively more valuable than the above.

EsEnt, AppFabric, IIS7 ARR, PowerShell cover your concerns above all of which are scriptable and automatable.

When you consolidate your time across the year spent on developing a backup methodology for your server, managing updates etc, monitoring, testing, where does that leave you? I know Linux inside out (I've been a professional user on and off since 1998) and I know you hit the same number of roadblocks as windows along the way.

Also your deployment model sounds odd. Do you check in your production artifacts/dlls and configuration? Do you build on your production boxes? Also your demonstration applications are tiny. A single deployment we throw out has 95Mb of release compiled DLLs across 48 assemblies...

We use NHibernate and have a test harness for MySQL already so we're not bothered about SQL pricing (anyway it's still plenty affordable as our product is valuable and pays for the licensing).

I was using rsync on Windows 2000 back in 2002 to sync press image libraries so that's a moot point really.

Azure cuts it for the ramp up phase of your product if you factor in the above. We did a proper cost analysis.

Are you seriously claiming that Linux is "not seen at all" in the "deep web" (in-house corporate development) ?

Hyperbole aside, obviously.

Or am I misunderstanding you?

BTW, if your experience is mostly in the UK then that might explain this viewpoint. It's much less true in the US in my experience.

>> I'd argue they are cumulatively more valuable than the above.

If you're trying to suggest Windows Enterprise applications have more cumulative value than everything else that's running on Linux than I think this is more telling of your limited world view than any kind of reality. Java ranks nearly 3x higher than .NET on the latest TIOBE Programming Community Index http://www.tiobe.com/index.php/content/paperinfo/tpci/index.... and that doesn't include all the other languages running on the JVM platform. In the Open Source space, .NET doesn't even make it on the top 10. How you've concluded Windows provides more cumulative value should provide great content for a MS Facts whitepaper.

>> EsEnt, AppFabric, IIS7 ARR, PowerShell cover your concerns above all of which are scriptable and automatable.

These options pales in comparison to what's available on Unix, which are much more popular and require much less tooling and are more scriptable. Big tooling falls down when you need to do something that's not supported out of the box.

Let's identify big companies that rely on Windows? and I'll come up with a few names that host on Linux. We can start with the worlds most valuable companies and work our way down.

>> Also your deployment model sounds odd. Do you check in your production artifacts/dlls and configuration? Do you build on your production boxes?

We use the repos as they are on Git, every project on Git is self-hosting either in NuGet /packages or contained in the top-level /lib folder (we're moving to NuGet for all future projects). We're currently building with xbuild on our servicestack.net server but as we've recently got Mono/Windows TeamCity CI Builds for ServiceStack's projects, we'll be switching to CI deployments. Also unless it's not perfectly clear, because we deploy from Git only the delta's are pulled in each deployment, so libs that don't change aren't pulled.

>> Azure cuts it for the ramp up phase of your product if you factor in the above. We did a proper cost analysis.

I'm still not buying it. Maybe it makes sense for a Windows-only .NET dev team, but factoring all costs Azure would still cost us at least 11x more.

Well.. I can state that Wells Fargo Bank, Intel, and American Express all have .Net development in-house... As do a lot of other fortune 500 organizations. In fact, I'd be willing to wager that 90% of the F500 list uses Windows, Linux, .Net, Java and other development. The pissing match is moot anyhow. I will say that I feel that Windows + .Net is a much friendlier environment to develop in/with than Java.

That said, the more I "get stuff done" with Node.js, the more I absolutely like it. Many small parts working together, integrating with other systems, and deployable on disparate platforms... It may not be "enterprise" and the tooling may not be there yet.. but doing with a 100-line script what takes a couple of projects in typical "enterprise" fashion is really nice. Our email spool handler is an 84 line node script. Our API service is 122 lines in Node.js (on top of some data migration scripts).

That said, Mono offers a lot... You can run your EXISTING applications on new platforms with minimal retooling, and reduced licensing costs... if you already have nix environments it's less costly all around. If you're new to nix, then sticking with .Net/Windows has other advantages.

Used in conjunction with other platforms is a far cry of being cumulatively more valuable then everything else put together. But I digress.

Agreed Mono does offer a lot of value, it's improved leaps and bounds since Xamarin's split from Novell, they're now well staffed and more focused on delivering their leading best-of-class products. I've witnessed tonnes of improvements to MonoDevelop since their focus on MonoTouch and OSX. From what I've seen at MonkeySpace Xamarin's got plenty more exciting announcements up their sleeve for next year.

With that said, it's making a lot more sense to depend on well supported Mono-compatible libraries so you always have an easy escape option out of the expensive Windows-only cloud provider lock-ins. You'll also have the option to use any of the exciting Linux cloud initiatives on the horizons like: Google Compute Engine / OpenStack.

Some comments:

1. Tiobe is a load of rubbish. Scraping the search engines doesn't give useful statistics - just a talking point. It also backs up my point about the dark web. There are huge swathes of professional software engineers who don't participate in blogs and the open source movement. Also lots of old stale information is out there clogging up the web.

2. Have you ever had to manipulate a heavily nested apache or nginx configuration file with sed/awk? (even with debian's nice sites-available/sites-enabled structure?). The only hope is to write a script to regenerate the files on demand. Chef/puppet make this easier but it's still a pain in the butt. In fact I'd say compared to appcmd, it's an absolute nightmare. I can think of many examples like this.

3. Name 20 companies that are Linux only. I can name 50 in the FTSE100 that are windows only. There is more market value in the companies below the top 100 than there are in the top 100 in most exchanges (do the stats).

4. Thanks for the build clarification. We use TeamCity as well and do integration builds from our VCS master but we publish from team city artefact drops, not from our VCS. We don't use NuGet as it only started to support dependency materialization recently.

I can name 50 in the FTSE100 that are windows only.

Define "windows only". I've worked at several large "windows only" companies, and one thing they all had in common was that lots of people all through the company where using Linux for all kinds of things.

>> 1. Tiobe is a load of rubbish...

Than pick your vendor neutral index. What non-MS sponsored stats shows Windows server deployments is greater than everything else combined? C# doesn't touch Java let alone anything else that gets deployed on Linux.

>> 2. Have you ever had to manipulate a heavily nested apache or nginx configuration file with sed/awk?

With sed/awk? No why would I need to? Especially for multiple sites / virtual hosts I've found nginx is a lot easier to configure than IIS. Isolated virtual host configs make it easy to maintain. Removing a site becomes as easy as taking down a sym link. Common tasks like Reverse proxying, Url rewriting are a synch.

>> 3. Name 20 companies that are Linux only. I can name 50 in the FTSE100 that are windows only.

Great, this might be our chance to finally get some facts out of you - Looking forward to your "50 in the FTSE100 that are windows only" list.

Here's a (more than 20) list of Linux-only adopters: http://en.wikipedia.org/wiki/List_of_Linux_adopters Big names include Google, CERN, DreamWorks, NYSE, London Stock Exchange,...

> list of Linux-only

They may be Linux users, but are far from being linux-only...

Re. 2: Have you ever tried to edit windows registry by using regular expressions on the file itself? No... because it's a silly idea to begin with / strawman argument. Store the data in some kind of data base and generate the configuration from templates. Better yet - use modules that just pull the configuration directly from somewhere (ldap for example).

To what level are we going with 3? Really all it takes is a single router / phone / storage appliance / ... and you're not windows only anymore. I guess it depends on your environment and preference too - I tend to run into companies which either don't see a reason to use Windows or use it only in some departments where it makes sense (support for example).

The linux / windows -only company is a strange concept these days anyway. Even Canonical is not a linux-only company in the end, since they have to make sure installer plays well with Windows and they have to develop Wubi.

I actually said your first point.

Re registry editing, it's easy and you can edit it in windows with any script based tools as the registry is a filesystem in powershell. Enumeration example - google the rest:

   PS C:\> cd hklm:
   PS HKLM:\> cd ./SOFTWARE/Microsoft/windows
   PS HKLM:\software\Microsoft\windows> gci

I use C# w/ Windows Server for game servers but I'm in the process of migrating to Linux. Why? Because everything else I have runs Linux so managing an extra OS is a lot of overhead. And Windows Server automation is hugely painful compared to Linux. With Linux automation for example, you can go from Ansibel which is a wrapper around SSH to Chef/Puppet that are very very powerful but much more complicated. With Windows, you get crappy Chef/Puppet support and the MS automation tools (from what I've seen) are too complicated for most people. Even setting up remote Powershell access is painful compared to setting up sshd. In short, there's monetary reasons for running Mono as well.

A UK .NET developer? A rarity on here... are you looking to hire anyone?

We're a pretty quiet bunch. There are lots of us - if not thousands. We're not hiring at the minute unfortunately. I've got your email address - will drop you an email if an opportunity comes up.

Can you e-mail me where you work - I'd just be interested to see what you're doing...

Really? I've seen quite a few on here, and there is a ton of .NET work out there in the UK. If anything, in Bristol we're crying out for developers.

Yes I know there's plenty of work in the UK but you don't find many .NET devs based in the UK on here.

Well, I can't move to Bristol and I don't want to work full time for anyone - are you hiring remote freelances and do you pay decent rates?

"You might be saving 11x on an Azure instance, but your ops time and staffing requirements will cost more than that."

Could you elaborate on why you think this is true? It seems to me like Microsoft baseless propaganda from last decade.

> We've gotten great up-time with Linux where our last server reached 480+ days up-time before we switched hosts

I'd worry about reliability and security of any service with that kind of uptime. Security if the uptime was achieved by not applying kernel updates, or reliability if some kind of hack to update a running kernel without a reboot was used.

The latter is a reliability problem because rebooting after a major update serves a purpose other than just loading the new kernel. It also serves the purpose of showing that the update has not broken configurations or dependencies required to boot. It's a heck of a lot better to discover that an update has left your system non-bootable right after the update, rather than finding out months later when some hardware failure takes down the server and then you find it doesn't boot and you've got to go back through months of changes trying to figure it out.

...I imagine that "showing that the update has not broken configurations or dependencies required to boot" would be addressed by having testing the configuration on an exact mirror of the production server, including rebooting, right? (of course, after a reboot the testing server is technically no longer a mirror of the production one...)

...of course, the OP should enlighten us about whether this is how they did things

Well, i don't agree here. You have the exact same system to test updates first, before rolling out on production! Just rebooting "because there may be something, let's see if it still works".. nahh, we're not running Windows here ;)

By the way, I have had servers running several years on the internet, without problems, one FreeBSD which topped all until some hardware maintenance was necessary.

I love Miguel et al. but I tried for a long time to make a public ASP.Net (MVC2) site work. It would run fine for a few days but inevitably would become unstable where the Mono Apache process would runaway and put the CPU beyond 100% usage (how is that even possible?). After a year of doing weekly restarts I couldn't take it anymore and moved back to Windows. The extra $30 / month in Windows licensing on Racksapce more than paid for itself in the IT headaches (and expenses) it saved.

Things may have since changed - this was back in v2.12 days - and / or other stacks (like service stack) may be more stable, but it's just easier for me to stick with Windows.

With all this said, however, I have had fun trying the MonoTouch toolkit demos. I hope to find some time to build a real iOS app (one that works with the aforementioned website, in fact). Fingers crossed things will work out better this time for me and Mono.

You're going to always have a better experience from Web Frameworks that officially support and have tested their stuff on Mono. We've added special normalizing support to match the behaviour of IIS/ASP.NET where we can, e.g. imitate redirects + case-insensitive file paths, etc. This reduces the differences when porting your ASP.NET app across to run on Linux.

Now that it's open source and MVC is accepting patches, MVC3 might be more stable on Mono.

I think the problem here specifically is mod-mono and the fastcgi server (basically the xsp-derived glue). I've seen a number of people complain about memory leaks and instability and having to restart apache. It's not readily reproducible but it's enough to make people too nervous to deploy.

Those sub-projects appear to be unmaintained and queries about their status on the mailing list go unanswered. Without some solid info there, it unfortunately makes the open source MVC stuff rather moot.

I concluded (perhaps wrongly?) that web wasn't a priority for mono and chose another technology for that part. On the other hand, mono makes a great socket server stack, and I still use it extensively there.