Hacker News new | past | comments | ask | show | jobs | submit login
Microsoft developer reveals Linux is now more used on Azure than Windows Server (zdnet.com)
426 points by Element_ 4 months ago | hide | past | web | favorite | 176 comments

Every few months we get another post like this, with a slightly different measure. Yes, lots of Azure workloads run Linux, and Microsoft makes it a priority. It's not news anymore.

At the same time, Azure itself runs largely on Windows servers in Service Fabric (https://github.com/Microsoft/service-fabric), so it's hard to count. With nested hypervisors as the normal architecture nowadays, this whole tally is pretty suspect.

But we (the tech community) never seem to get bored with the "Microsoft is eating humble pie on OSS" story. Maybe that's exactly what Microsoft deserves after so many years of big-bad-wolf behavior. Unfortunately it makes their real engineering and OSS feats hard for us to see or acknowledge.

Like Service fabric. Did you know that Microsoft built and open sourced an orchestrator that can manage tens if millions of nodes with totally diverse workloads (containers, VMs, and bare metal), nested hypervisors, and even nested orchestrators in a secure multi tenant context? That beats the crap out of kubernetes on a capabilities front, but it's not an easy story to sell. Or that Windows Hyper-V has done thin-VM/M container isolation on par with kata and firecracker for years? Or that many (if not most) Azure services are open source? You can run the entire azure stack on your home server closet if you like, especially including their biggest growth areas of kubernetes and ML/cognitive services. If you want an open source alternative to AWS, azure can be a real option.

I enjoy the historical irony of this as much as the next guy, but we really have to stop letting it blind us to the awesome things MS is doing that don't fit that narrative. Microsoft is benefiting mightily by adopting open source culture, code, and practices as fast as it can. And if we can get our heads out of our own self-righteous asses enough to upvote the stories, we can profit from it too.

>But we (the tech community) never seem to get bored with the "Microsoft is eating humble pie on OSS" story. Maybe that's exactly what Microsoft deserves after so many years of big-bad-wolf behavior.

And Microsoft absolutely relishes it. The tech community provides priceless support for Microsoft's key marketing message, which is that Azure runs everything and you don't have to go to AWS for first class Linux hosting.

The humble pie they are eating is worth billions per bite.

And the next generation at Microsoft doesn’t even think this is humble pie - we think Linux (and open source in general) is the most delicious (and profitable!) pie we could possibly imagine.

> The humble pie they are eating is worth billions per bite.

Absolutely it is - I don't mean to deny that. Only that the story obscures a lot of value that we, the broader tech community, should be taking from their contributions.

I've been a dev for 20+ years now. I don't care about the OSS vs Proprietary argument at all. I care about my productivity and job satisfaction. Microsoft - in every sphere from IDEs to cloud platforms and now machine learning - has done more to make this stuff easier to learn, use and afford than any other organization.

>I don't care about the OSS vs Proprietary argument at all. I care about my productivity and job satisfaction.

That's kind of like saying, "I don't care about voting, I just want politicians to represent me."

Arguing on the internet isn't analogous to voting. Voting accomplishes things, arguing about OSS Vs Proprietary just wastes time.

It matters when you're involved in purchasing decisions and also when you're deciding whether or not to contribute to a project.

Well Oracle...is a counterexample to that argument.

I think it's saying that "software politics" and productivity are not one in the same.

I would give that title to SUN. Without them, there would be no Java, and thus no Eclipse, and thus no incentive to advance the IDEs to the point they are today.

That productivity and job satisfaction, you have it despite Microsoft. Imagine a world where MS wouldn't have spent their resources on killing Webbrowsers and Java, where you could run a jar file on any operating system with ease. Or imagine a world where exchanging office files between different programs would have been easy.

> Without them, there would be no Java, and thus no Eclipse, and thus no incentive to advance the IDEs to the point they are today.

I don't buy this argument. IDEs would just advance as they're tools written buy the people who use them - there's an inherent drive to increase their usefulness and automate the mundane and repetitive. Also for what It's worth I think before MS really got their shit together in the IDE space it was Borland that was at the forefront.

Of cource, IDEs advance. But at which pace? The bottleneck was the language. When MS got their shit together they just caught up with Java.

Maybe MS has done the most, made the biggest investments. But splitting the ecosystem of managed languages into Java and C# hasn't made life easier for developers nor users.

Which orchestrator are you referring to? System Center in no way competes with kubernetes since it's lacking 90% of the features. Maybe I'm bad at googling, but I can't find what you're referring to.

Edit: I see the one you're referring to in that link, but again, it's not at all a comparison with kubernetes. It's easy at face value to say this one is better, but it's missing the point of kubernetes, which is extensibility, following standards, etc.

They linked it. Service Fabric.

> Service Fabric is a distributed systems platform for packaging, deploying, and managing stateless and stateful distributed applications and containers at large scale.


If you Google it you'll find several comparisons to k8s and others.

I looked into this a while back, and Service Fabric was a clear winner over k8s, and resource use was a fraction of k8s too.

But, I felt it needed to gain community traction before I could commit to using it - knowing my luck, I'd build something and Microsoft would abandon it because k8s...

Kind of a chicken and egg problem, which is a shame really.

No, it was a clear winner for your use case. It's missing a significant amount of features from k8s, though, and if you don't need those, then great. But don't pretend they're comparable.

I think it's more fair to say the suit different use cases (an MS person on HN explicitly said this last year), rather than competing.

As I remember it, it was a clear winner with no caveats. But my memory is shit, so there is that.

What features do you believe it's missing?

Custom scheduler, a declarative API, the estimated of custom resources, many different filesystems natively supported, etc

By "scheduler", do you mean as in "decides which nodes to run on"?

You can build Service Fabric manifests with XML.

Yes, but from what I can tell in their documentation you cannot make the same level of decisions that you can with kubernetes.

RE your edit, Service Fabric came out well before any standards existed.

>following standards

What standards?


" Did you know that Microsoft built and open sourced an orchestrator that can manage tens if millions of nodes with totally diverse workloads (containers, VMs, and bare metal), nested hypervisors, and even nested orchestrators in a secure multi tenant context? That beats the crap out of kubernetes on a capabilities front, but it's not an easy story to sell."

Interesting .. Can you please elaborate more on this ?

You are trying to defend Microsoft from these stories. But "Microsoft eats humble pie on OSS" is exactly the narrative they want out there. It gives them the cover they want to embrace extend extinguish Linux and other major open source projects.

> It gives them the cover they want to embrace extend extinguish Linux and other major open source projects.

Were you alive when Microsoft reigned as the dark lord? The contrast between life then and now is so dramatic. They simply lack the power they once wielded over the world of computers and the internet - to everyone’s great benefit, including theirs.

>Were you alive when Microsoft reigned as the dark lord?

Yes I was. I worked in that stifling environment when anything not MS related was dismissed out of hand. Where MS's own marketing exhorted everybody to replace their 'legacy' unix systems with e.g. the wonderful system of Biztalk, Exchange and IIS.

Yes, constant vigilance so that they never regain that position again, as I'm pretty sure they desperately want to.

> Yes, constant vigilance so that they never regain that position again, as I'm pretty sure they desperately want to.

I doubt Microsoft wants to reign as a dark lord again at the moment - rather, they're attempting to grow their business through use of open technologies, like Google back in the 2000's. Although if any MS products/services become too dominant, I have no doubt that (like any other big company in such a position) they'll fall to the temptation to exploit their dominance for more money and power.

It's not Microsoft that's evil -- it's people and their capacity to be corrupted by inordinate power.

Lacking power is not to their benefit. You have drunken the coolaid all the way.

Personal attacks will get you banned here. Would you please review https://news.ycombinator.com/newsguidelines.html? It would be good if you'd stop posting unsubstantive comments generally.

> Did you know that Microsoft built and open sourced an orchestrator that can manage tens if millions of nodes with totally diverse workloads (containers, VMs, and bare metal), nested hypervisors, and even nested orchestrators in a secure multi tenant context?

No, and I don't really care. Is there something wrong with that?

Am I limiting myself by ruling them out from the get go based on past behaviour?

After a decade of out right ignorance of windows services what would it take for me to catch up in order to even begin to unpack and then finally even begin to leverage what you just mentioned?

Is it even worth the time for me to try and grok it?

Am I laughing that azure runs more Linux than windows? No. I assumed that was the case years ago because you're an idiot if you're still stucking on that windows teet after all they've done. And I don't care that you think Im a close minded idiot for speaking that truth.

Why? Because not a single employer that matters to me thinks your knowledge of windows crapware is a competitive advantage in the environments most real tech operates in.

Call me when Google starts running search off of windows boxes.

> Am I limiting myself by ruling them out from the get go based on past behaviour?

Yes, by definition. You have every right to, of course! But yeah you're limiting yourself.

> what would it take for me to catch up in order to even begin to unpack and then finally even begin to leverage what you just mentioned?

What did it take for you to unpack/leverage k8s? Similar. SF runs on Linux, and is an orchestrator for fundamental components that you're used to: VMs, containers, and bare metal instances. There's nothing Windows-specific there. Particularly if you deal with a mix of stateless and stateful applications, or if multi-tenancy is an issue for you, SF has valuable strengths.

> not a single employer that matters to me thinks your knowledge of windows crapware is a competitive advantage in the environments most real tech operates in.

Hm, the only windows specific component I mentioned is Hyper-V. Other services tend to be solid implementations of open source projects you probably tolerate, like kubernetes or tensorflow. You know, places where MS is a major contributor. I was talking about our blindness to MS contributions across the spectrum - which you seem to exemplify pretty neatly.

You can limit yourself however you want. Be careful with your blanket statements though; you just excluded a lot of companies from your list of "companies you care about" and "real tech." Airbus, Atlassian, Bitnami, Cisco, Dell, Docker, Facebook, Fujitsu, GE, Hashicorp, Intel, Kaltura, Lockheed, MongoDB, New Relic, Oracle, Nokia, Raytheon, Red Hat, Roku, Salesforce, Samsung, SAP, Siemens, Sony, SUSE, Symantec, Tableau, Trend Micro, and others all work closely with Azure and Microsoft tech... maybe those companies don't count as "real tech" to you. Either that, or you should consider re-evaluating your limitations.

You haven't mentioned how I am limiting myself.

How am I missing out?

Most if not all of those companies are uninteresting to me.

And I don't think any of them would shy away from hiring me for my lack of experience with azure / ms tech. They might not hire me for other reasons but that would not be one of the reasons.

I see a lot of government shops picking Azure because nobody ever got fired for buying Microsoft.

But once they open up to the cloud I see a lot of people installing stuff like Shiney servers, Jupytr notebooks and other open source software on Linux servers that would never have been a starter in an enterprise environment.

The government department standardizes on Azure but all the young devs push for Linux servers so all their shiny toys work great.

Windows is truly dead on the cloud and has been for a decade. The only thing people use it for is self hosted exchange, and file servers and active domains - legacy stuff.

There is an entire parallel community of business IT, basically unknown and unintelligible to the Silicon Valley community, where the Microsoft server-side stack is as obvious a choice as Linux is in our world. You may even have some of it in your company. It's really amusing (and dysfunctional) when our projects require "Eng" and "Corp IT" to cooperate. We're both ostensibly computer professionals who select, write, and operate production services, yet to each side the other may as well be from Mars.

And at least where I come from, it seems the owners of .Net and C# shops are all driving Maseratis or Mercs, while the owners of Ruby/Nodejs shops are on fixies or electric skate boards.

(And the Java people all work for banks, and ride the train. Even though they're really well paid, they don't own the company, and don't get allocated parking spots in the bank's CBD offices...)

Probably an awful statistical analysis, but it's a common enough thing here I've heard other people talking about it...

Like 10 years ago, I thought it was great when we garnered a client which was one those "Microsoft only" type shops, because I like C#/NET. But the only justification for using it was really "IT made us do it". And I saw how the IT leadership in these places were older people perhaps maybe fighting older political battles. In any case its been many years since I've stumbled across a pure Microsoft shop. I think Azure means they know they lost, but also have legacy business to support.

I work in a 99% MS shop, basically (well, apart from our SAP installations). Most of this was indeed setup like this due to homogenization (all our admins know everything, it all works together, no weird stuff on some servers, etc), but it has actually created a weird situation where we have some smaller services running on old unix servers and no real admin in the org... we actually have on older guy coming in once a week to check on these servers. If it wasn't so sad it was hilarious. I mean, I used to do some unix server stuff en years ago, but here I am an SAP dude, so while I do have some auths on one of those servers, I can probably basically restart some daemons and that's it.

Oh yeah, and iPhones. We have those - not complaining, really, but those were sold internally as "it looks better if we have those premium phones and not old junk". Ah well...

>But the only justification for using it was really "IT made us do it". And I saw how the IT leadership in these places were older people perhaps maybe fighting older political battles.

I've seen this happen too in predominantly Linux shops. In the early 2000s Microsoft worked hard to cultivate relationships with future IT leaders and those relationships have a way of enduring. Even if your CIO can't get everyone onto Windows they might still find a place for Office 365 without much time spent weighing up the alternatives.

That’s really how it works. I can attest to that as a former IT guy in a large organization. It’s a personal relationship with a clueless CTO. But to be honest, they’ve just replace IBM guys doing the same.

Big difference, IBM representatives actually knew their stuff. Microsoft would rely on its enterprise sales guys to cultivate this relationships, pushing CTO to make the most stupid decisions and using system engineers to sugar coat and hide the details.

I remember one being told off by a sales person when saying “yes, it’s a bug” in one of these sales pitches.

IT made them do it for good reasons though. Managing a large number of users and workstations in an enterprise environment is one of the things that Microsoft stack stands out.

Pull the boundaries in. Provide all of the enterprise services through the web. Provide hardened VMs for those services that can't be migrated. Suddenly your workstations are thin clients, and it doesn't really matter anymore if someone decides they want BYOD.

Now you can get after the actual interesting problem of intellectual property theft and corporate espionage. Things which the firewalls of old did nothing to prevent but provide a layer of obfuscation.

It's naive to think that just because your production critical software is web based, you can just have everyone BYOD. That is asking for rampant malware on the company network, and targeted attacks that steals company secrets.

You're completely missing the point. The internal network I'm describing should be nothing more than server metal and virtual workstations. Your border is miniscule at that point. You no longer have to fret about network ports in conference rooms. You no longer have to worry about testing your OS and software patches on the 20 iterations of lifecycle replacement hardware that is floating out in your company. You don't even need to invest heavily in workstations, as thin clients will suffice for most situations. Business continuity becomes a piece of cake because you're no longer factoring in the physical workstation beyond ensuring your cold site has a box of laptops in the corner.

Your concern about rampant malware won't even matter because the only way people can access the web services is through the VMs. The thin clients and physical workstations won't even be able to access most of those services that are mission critical.

In the situation where a company provided physical workstation is necessary, that machine would be just as isolated from the internal network as anyone else. Developers can use one of those, or use VMs that are VLAN'd off. And if your developers are automating their builds and containerizing, then your developers are going through layers of services and automation before their code goes out to production, so even they won't need access to production environments from their relatively insecure dev laptops.

From the lens of a programmer it's definitely possible to do BYOD without any real danger to the domain.

They need access to a few webfrontends and be able to use their SCM server...

But in order to do that, they'll have to be able to test locally or have a really hardened access to their servers. Significantly harder than just forcing everyone on company hardware without root/admin access

Developers get special treatment, because the nature of their work often require them to have local admin access. But the majority of people do not need that, and if I'm in charge of company network and security, I would not allow anything but approved machines on the internal network.

> Developers get special treatment, because the nature of their work often require them to have local admin access.

That is extremely rare at larger b2b enterprises

Each and every organization I've heard of that did that, reverted to give developers local admin pretty quickly because the requests to the IT administrators for every time a developer needed admin to install a required dependency, start whatever at admin to debug, change reg settings, test installations, ect.

Senior scientists in corporate R&D often require local admin access as well. The image analysis and spectum simulation programs I needed to run and keep updated were not on the radar for our corporate IT folks. They were very good at what they did and were cooperative when we explained our needs and showed that we were competent to manage our own systems and were willing to reach out and ask questions before doing something unfamiliar. Even in R&D, most chemists could use a standard workstation. It was mainly analytical chemists/material scientists working with specialized instruments and doing custom software development that needed local admin access.

“The company network” is no different from the coffee shop WiFi next door, just has more bandwidth.

We still restrict web access to company managed machines but it’s a layer, not an essential boundary.

Provide all of the enterprise services through the web.

Thats like 10 years of projects right there

I work in a mostly MS shop currently. Of course, all the orchestration scripts I've written for CI/CD etc around our projects are using Docker and Node. Then again, in the main project I'm on, everything I didn't make is .Net Framework and not Core... so going to spend months migrating the crap to actually make progress with some other needs.

There are some things I do like about C# and .Net ... the culture is rarely one of them. Though I still prefer it to Java for the most part.

Basically in the same boat. If you haven’t looked recently, the container support in Windows is getting much better - including prod support for Windows worker nodes in Kubernetes 1.14. But just using Docker with gMSAs was a big help to our CI/CD and simplifying/standardizing deployment of our apps which are a mix of legacy ASP.NET, WCF, and traditional Windows services.

If you can target at least 4.7.1 you can use configuration builders [0] to use environment variables/json/cloud parameter stores/etc to modify legacy web.config or appsettings without code changes. You can also use the .Net core config libraries directly in legacy .Net apps if the devs are willing.

[0] https://docs.microsoft.com/en-us/aspnet/config-builder

I had a coworker who came from Facebook and told me similar stories.

We all hear about the exciting engineering projects and stories coming out of Facebook. But the IT side of their org is held together with various off the shelf enterprise products and Windows stuff.

Everyone uses Active Directory because its basically the only functional solution for its problem space.

Which problem space is that? For group policy part cfengine, salt, puppet are far better (group policy is just modifying windows registry and manipulating files although at first it seemes a bit like magic). For authentication we have kerberos, which Microsoft pulled in from the free world beginning with Windows 2000. Kerberos does not fly in the Cloud world though.

And Dropbox is just sshfs, ftp-mount etc. There is a huge advantage to having an all-in-one, it just works, solution. Active directory is such a solution and does really well. Login to a different mschine and things are just setup. Push programs, setting, updates. Change security. It's great.

Have you ever actually done that - managed a large fleet of desktops and a productivity suite, all on Linux with no active directory? Complete with whatever the Kerberos equivalent of a backup DC is?

The problem with AD is that it is for windows only. It does not fly for managing Mac and Linux desktops. So that gives us salt for example, which has clients for all three. The second problem with ad is that if you need to manage any complex settings, you'll need to write your own templates. The included group policies are only for basic level os management + some for basic level Microsoft office managment stuff. Anything else and it's scripting + manual work and AD is only for distribution and selecting hosts/users where to apply those settings. And yes, I've managed multi thousand workstation networks with AD. Do not recommend it.

AD isn't just for Windows, which would be weird since it is mostly a fancy key value store (with associated functions and services of course). SSSD for example can use AD. The problem is that Linux itself doesn't support the same functionality client side, which using a configuration manager doesn't really solve. And question wasn't if you have used AD, but if you have managed Linux desktop deployments without it. Since your claim is that it is better.

I'm in the process of bulding a solution for managing all three OS'es. AD is not on the table because theres nothing to do with kerberos in our network and AD would be a "windows only" solution.

Why is AD a Windows only solution? Large corporations and startups use it to run tens of thousands of Macs and Linux machines in addition to Windows. In fact, I can't think of a single large company that does not use it. Its basically the core for many.

Linux can totally run in a AD domain with auth managed by AD. Client side SMB is also not bad. But you are excluding Kerberos for some unrelated reason, right?

I would imagine that Google has and is doing that, as is AWS.

You'd be surprised how big of a Microsoft shop Amazon is. It's pretty representative of a large company actually.

Hosting SQL Server has been another common use case for Windows Server for a long time, though it can run on Linux too now.

Last I looked many of SQL server advanced features didn't work on Linux. https://docs.microsoft.com/en-us/sql/linux/sql-server-linux-...

Right, like no Analysis & Reporting Services on Linux means if you're using it for BI, you're probably still running it on Windows.

The last time I installed sql server was on Linux

Government does it because Microsoft was genius in sneaking cloud service terms back in 2011/2012 beyond the scope of O365 into the Microsoft EA product use rights before government lawyers and procurement people figured it out. O365 had ROI so it just rolled through.

They have a leg up because they have been available to every government entity for almost a decade. AWS and others are still hard to buy for many government entities.

On Azure, a Windows VM instance tends to cost about 50% more than the equivalent instance running Linux,so it is a no brainer to use Linux if your application is operation system independent.

If I encounter an issue on Linux (that I can reproduce), I can investigate it to its root (sometimes in the kernel driver) and fix it (or find someone who has fixed it). On windows, I have only seen issues fixed either by reinstalling applications (mostly randomly) and praying or by using annoying work arounds. IMHO, the cost is higher in Windows.

This is to some extent only a consumer approach. There are tools allowing in depth debugging. Given a big enough budget you'll also get direct help from MS. This relies on very skilled staff though.

Sure, it's easier to debug kernel things on Linux. But you certainly can debug Windows as well.

Stuff suddenly starting to work on reboot for Windows is getting old.

I don't know what to use Windows servers for if it wasn't for a project that uses MetaTrader.

We have these old Windows admins now learning to work with Linux. What drives me insane is their habit of randomly rebooting machines if something isn't working right. There is a problem with a Linux box, they ask you to take a look at it, you SSH in, type "uptime"... bam, it's two and a half minutes. A lot of precious diagnostic information has been lost.

This Reboot-to-Fix mindset encroaches on other domains as well. My car had a weird issue when it would't pair to the phone, auto-technician: "reboot!" Someone here say that I'm probably a *nix admin vs a Windows user, and that's true, but every time I try to do adminsy things on a Windows box, the tools are just not there, or the interface to the tool is like controls in an alien spaceship. I find myself completely and hopelessly lost. Trying to find stderr, stdout, logging, grepping, finding, the tool to work with PKI certs, all fearsome in their inscrutability. Are there Windows kernel old-timers who can help me with the learning curve? No. Most Windows admins don't have a habit of digging deep because that's not the culture, because the tools weren't/aren't there. And reboot fixes most things on Windows.

alias reboot='echo "This is not Windows!"'

Real Linux admins know how to invoke the real 'reboot' command.

Make sure to set SSH to log in directory to vim. Unlike Real Linux admins know how to quit vim so it will be no issue for them.

Wouldn't their login session end after quitting vim?

Yes, but you can get to a shell without quitting vim. <esc>:sh

Oh please, they started Gnome and used the shutdown command from the menu.

init 6

With the growing ubiquity of the linux port of svchost.exe(systemd), nested hypervisors, containerization, et al. Rebooting (where feasible) has already become a front line option.

This is a roundabout way of saying: "I know more about Linux's internals than Windows'." Essentially you're a power user of Linux and a consumer of Windows. I'm not sure that's Windows' fault.

That's not entirely true...

Given Linux's history as OSS, it has historically been easier to be familiar with the inner workings of Linux than Windows simply by having access to the code. Other than reading System Internals, becoming MCSE certified, etc. it was really hard to know much about Windows. I'm not saying every Linux user has read the source but being built on top of OSS does allow the user to dig deeper if one wanted to. I say this as someone who started out his career in Windows and Microsoft and then later went to Linux (because I went from corporate to startup). It's much easier to pinpoint a problem with OSS and Linux, especially in situation when the documentation is lacking. It took me a lot less time to have a decent knowledge of Linux and OSS systems than it was for me to know the equivalent on Windows and Microsoft.

This effect gets compounded when you consider an entire ecosystem.

Yeah, I mean why doesn't everybody dig through the kernel sources on Windows when they encounter a bug like that?

The vast majority of Linux users will never poke around the source to the Kernel ferreting out a bug, but they do have the option. On Windows it's much harder to get access without being a Microsoft employee.

It does not end with the kernel. Of course it is possible to run proprietary software on Linux and OSS software on Windows, but I have rarely had to deal with proprietary software in Linux. That is only in small part because I avoid it.

I'm not a kernel hacker. But I have on occasion cracked open the source for some product on which I relied to find out why it was behaving contrary to my expections. Much more I have done that with a library upon which I was relying.

Many many times more often I have sought answers from authors or experts of a lib, etc. to clarify what I didn't understand from documentation.

All of this is easier and frendlier with open systems. MS is coloring itself more open recently, but they and the ecosystem built upon them and around them have a long way to catch up. From my perspective the gap is so large and so pervasive that I don't even bother kicking the tires any more. I don't know if MS will make it there in my productive lifetime. If they do, I think the ground will have to have shaken dramatically so many times that it will get my attention. So, I will continue to ignore them until then.

I recently went through the Azure Fundamentals certification for work and one of the questions on the test (written/approved by Microsoft) even revolved around how to get a VM spun up as cheaply as possible, and the correct answer was "use Linux".

The new Microsoft doesn't care if you use Windows or Linux in Azure as long as you're using it in Azure. Windows licensing is far less profitable than any of the SaaS or PaaS offerings they'd rather you use.

This is a side note, maybe it's been covered before. Has anyone noticed that microsoft.com runs on Apache now?


Years ago, Microsoft used Akamai as their CDN. Akamai ran on Linux/Apache.

Don’t know if that’s still the case though.

Why would they bother paying nginx the outrageous price for their server?

Or use iis

Or use Kestrel, which should be able to be front facing. It does very well with speed.

> "major cloud provider uses linux more than windows"

Well, that's to be expected. Sorry, this is in the same category as "Huawei CTO uses iPhones" or "amount X of Google employees use Surface Pro". I don't quite get what the article is trying to tell me.

Those are extremely poor analogies. You listed two things entirely irrelevant to a business as comparisons to a key component to a business.

"Satya Nadella once used a linux computer" is the category of thing you listed.

"Apple building android phones" would be a better snalogy for this.

> "Apple building android phones" would be a better snalogy for this.

Not really. Apple selling Android phones at Apple Stores is probably the best analogy.

Apple opening an electronics market and selling washing machines, iPhones and Android Smartphones and a lot of other things.

In essence it's all "Company/Organization X buys/sells/makes/uses stuff/services for/from Company/Organization Y". Well, that's what's business is all about. Overlapping market shares and interests included. This article is nothing but based on the old "Windows VS Linux" discussion. So yeah, I think it's in the same category as "Satya Nadella once used a linux computer" category and fits very well.

Windows VMs in Azure are significantly more expensive than Linux VMs because Microsoft still charges for the licenses. Only if your organization has already purchased a lot of them it can use the Hybrid benefit to get similar, but still higher prices.

Seems like a bad strategy for Windows Server in the long run, but they may have decided that it is already dead.

I am on one hand glad that Linux is being adopted, but I am mostly sad.

We only have two operating systems now Windows and Linux. (MacOS for the desktop as well).

It becomes a mono culture, and it takes away healthy competition and innovation.

I want many more operating systems, not fewer.

I with MS would make it free to run Windows Server, (something they move towards with the CLI only version). In order to hopefully increase adoptation.

I dont think that will happen as long as MS has a lot of enterprise custoemrs who are paying huge for the Server licenses.

I want more "kinds" of operating systems. It is ripe time to look forward to building ones not based on Unix, files, forks and execs. I want to explore different ideas which were not tested because Unix model became popular.

There is a lot to explore on the side of Smalltalk machines, Oberon, Lisp Machines of old. As a developer, live programming language based environments is my kind of thing.

Sure, but an OS that can't run a browser (i.e. it's difficult to port to) is going to have a massive adoption hurdle.

I was a big fan of the QNX demo floppy (1.44MB OS with browser), but that didn't lead to me using it on the desktop. http://toastytech.com/guis/qnxdemo.html

It's too bad that web browsers are so complicated and expensive to port.

There might be room for something new and different that runs on it's own VM (like Smalltalk) and can use libraries on the host to get things that do not make sense re-implementing, like a web browser. If it gains sufficient traction, the work to move it to a more stripped down environment would be clearly worth it.

You might like https://github.com/froggey/Mezzano, Common Lisp operating system.

You forgot to mention TempleOS :) Made from scratch (incl language & compiler), main interface is a shell in the same language the OS is written, designed for recreational programming...

> We only have two operating systems now Windows and Linux.

That all depends on what platform(s) you're using. For example, IBM still gives you zOS, AIX, the System i OS, and so on. And there are other operating systems for other platforms.

The computing world doesn't begin and end with Wintel, you know; it may seem that way sometimes, though.

AIX is very unixy, though zOS looks pretty alien from it's description. I would love to work on zOS, but there is very little window of opportunity for that.

I would love to see something like https://pharo.org/ booting directly. Its an OS for all intents and purposes (To my realization, languages like Smalltalk, Lisps, Forths ARE Operating Systems)

Yes, but OS is not like a javascript webapp. It is terribly difficult work to do and debug. And there are many other kernels and OSes already. Most kernels don't support the HW you need though. Even Google is writing Fuchsia!

Microsoftie here, a lot of Microsoft's own infrastructure is hosted on Windows Server instances.

Won't comment further than that, So Windows for server scenarios is certainly not dead.

I don’t see how the article says that “windows server is dead”. And Microsoft infrastructure is obviously running on windows, I don’t think anybody expects otherwise.

People would misconstrue the increased usage of Linux VMs on Azure to assume Windows Server instances are no longer used for Server scenarios, which is not the case.

> a lot

I see what you did there ;)

My favorite part about Azure is that Microsoft shops are totally fine with spinning up a Debian install simply because the image is available.

Prediction for the future: Microsoft will open source Windows but sell commercial support packages for it.

Windows server will be offered as a compatibility layer in Linux on top of Docker.

On one end, I really like the open source movement;

Having said that, I can't do anything but blame it for the slow pace of computer systems research. My reasoning is that the OS does not sell anymore (partly due to FOSS movement) and thus research computing environments hardly come out of the labs for use in production.

We are forever stuck with Unix's interpretation of what an OS could be, and there is very little compulsion/motivation/incentive (can't think of the correct word) to explore. People who do explore (example at MS research or Later Bell labs, remains in the labs)

This is not a new phenomenon, though; I was considering an OS research PhD all the way back in 2000, in the research group that brought out Nemesis ( https://en.wikipedia.org/wiki/Nemesis_(operating_system) ) - but even back then I could see that the compatibility issue was a colossal barrier. Back then we thought Windows had won the desktop and the UNIX-like OSs would remain in a niche.

The only thing that disrupted that was mobile. First Symbian, then iOS and Android. Sufficiently different use case that backwards compatibility was rendered irrelevant.

I guess thus is the same kind of evilution that s lotnpf technology goes through as it is adopted. Eventually, standards evolve (lane sizes, driving sides, electrical voltages and current limits etc...) and because they are adopted so widely, any change becomes almost impossible. C and UNIX look like they are going to be that standard for the computing world for the next few centuries.

That's very unlikely atleast in the near future due to things like https://devblogs.microsoft.com/oldnewthing/?p=98755

Microsoft should do an Apple and build the Windows 13 on GNU/Linux.

I also prefer a GNU userland, but Dave Cutler and his team did a pretty serious job with the NT kernel.

A post from a blog I like a lot about how Linux and NT aren’t so different: https://manybutfinite.com/post/how-the-kernel-manages-your-m....

Windows’ driver ecosystem is important to Windows’ market success - switching out the kernel (and so, the driver interfaces) would throw that all away.

.. unless they build an adapter.

If this sounds horrifying, consider that Windows already has a subsystem for allowing graphics drivers to crash and restart seperately. The reverse also exists: https://wiki.debian.org/NdisWrapper

In my eyes that would be a demotion of technology. They should instead do a Xerox Parc and invent newer/different kinds of hardware operating environments (and try to ship them too).

Even if they would, they couldn't do with Linux what Apple did with BSD because the licensing situation is different: BSD allowed them to develop a proprietary OS, GPL doesn't.

I've said this before, but takes like this have it completely backwards. The NT kernel is really good, that part should stay. Win32 is the source of most of the shittiness that people have come to associate with Windows, it should be replaced with a Linux userspace... which is exactly what WSL does (or at least what WSL1 did, now things are a little different)

I don't know if you're already aware of this, but macOS is not based on GNU/Linux. It is based on NeXTSTEP which was in turn based on a combination of Mach and BSD.

Also known as doing a Novell - if you want an example that didn't turn out so well for the company.

Please release the Linux-flavoured Windows desktop OS Microsoft.

It's the default choice when you set up a VM.

I'm surprised it took this long. With Linux support for .NET and SQL Server, there is zero reason to host anything new on Windows now (of course legacy enterprise software is another story). I wouldn't be surprised if Windows Server is fully EOL'd in a few years.

I can understand that for many Mac OS users it's hard to understand Windows. When your OS vendor is requiring you to rewrite parts of your applications each year because of changing APIs it's impossible to imagine a world where applications from 1995 are still running on operating systems from 2019. Same.with security updates for only 3 years where Windows gives you over a decade time to upgrade.

But when you are a manufacturing company and just need to get your product out, you don't care about your software. You want it running for as cheap as possible. That's where Windows shines. Mac OS is explicitly unfit here. Same with the hip technologies like nodejs that require constant attention.

Everybody gives Microsoft shit for moving slow but when was the last time you had to proof that your patch doesn't break compatibility with 20 year old software that you can't even get anywhere anymore?

Didn't microsoft ditch QA and make several broken patches recently? I was under impression testing new update rollout on your devices is still local admin responsibility and can break things. How is it different that staggered rollout of any other os update?

Yeah, they've completely borked the Desktop side of things, it's true. However the server side is much more conservative.

So, like OP said, legacy software that no one wants to maintain?

The comparison here is Linux, not macOS.

I think you underestimate how slowly things move in some places. There are still shops running COBOL on mainframes. Windows is going to have a place on servers for decades to come.

With .NET 5, Windows to Linux migrations are going to become faster.

One can only hope... Containerization on Linux is so much smoother imho. Not being able to run both win and lin containers for developers is a big issue. Linux owns this space, and it will likely continue down that path. Currently doing a couple .Net core projects with deployments to Windows and Linux.

News flash, there is more COBOL being written today that in any prior year. It's a small part of the total output, but is not decreasing in absolute terms.

What is the source?

Yes, but in the same way companies are still running XP. There is zero chance there will be new releases of Windows Server a decade from now. Microsoft itself is starting to host its services on Linux.

I don't see a Linux version of Exchange or Active Directory yet. For internal email, it's pretty much Exchange or GTFO.

OpenLDAP, Postfix/Dovecot?

Not a fully integrated email and calendaring solution that can be set up and administrated by typical overworked corporate IT staff. Completely unacceptable.

Then what about iRedMail? Isn't that sort of an all-in-one wrapper around Postfix/Dovecot and some other stuff?

And all the other stuff?

It's been less than a year since the last release of OS/2, I think we're a long ways away from the last release of Windows for servers.


Your statement reminds me of a well known tech pundit (Cringely?) who predicted in 2012 that Windows client would die in 2017.

You completely overlooked Active Directory. There is no open source alternative.

Who wants to deal with licensing headaches on cloud servers? Plus if you want to run headless then you need to go through the AD setup process to manage Windows hosts, where the Linux hosts are fine with just ssh access.

It's always fun to watch opinion changes to where the money is. Of course it's not about freedom, community or niceness. Their revenue from cloud services is growing fast. :)

I wish MS would just create a Linux or BSD distro like Amazon Linux, ClearLinux, Darwin etc. to replace Windows Server outright...

> to replace Windows Server outright

Why would they be replacing a product that earns them money?

> Why would they be replacing a product that earns them money?

Because they can make just as much money licensing versions of SQL Server et al for Linux or selling Linux cloud instances, not have to spend any more money developing a separate server OS, and make their customers happy because they can have Active Directory without Windows Update.

Windows Server is based on Windows Desktop, even the teams they kinda unified a while ago. Is not like they spend that many money directly on Windows Server.

they can make just as much money

I'm doubtful about that.

They could probably partner with Red Hat. I would've laughed at the idea 10 years ago, but these days, I wouldn't be that surprised. When they've had OS-ish integrations needed for things like machine-based authentication, RHEL has been treated like a first class citizen. I'm not sure how much Amazon Linux really changes other than branding and some baked-in tuning - unless you're really going to commit to maintaining the OS, just providing a minimal RHEL-based image makes a lot of sense.

Azure and RH are partners already, at least regarding OpenShift. With RHs push of RHEL Atomic and the purchase of CoreOS you also get minimal images of linux there. Another noteworthy thing are the Universal Base Images (UBI) by RedHat.

Why would IBM undermine its own cloud like that though?

IBM is not a serious player in the cloud segment.

Did you forget that IBM bought Red Hat? It's one of the strongest cloud players now.

I did not however, RedHat is not even remotely close to be one of a strongest players. IBM is hoovering around 10%ish, RedHat has no cloud offering that could be taken seriously. Most of the cloud vendors are competing in the public cloud segment (even though with VPC this is arguably more complex question) while RedHat focuses on private cloud, offering mostly OpenStack.

They have many, many partnerships that undermine their own offerings. Their strategy is less about being the best at something but being your one-stop-shop to get something that works for everything.

check out their kernel for WSL: https://github.com/microsoft/WSL2-Linux-Kernel.

Doesn’t WSL depend on a Windows host though? Can WSL be used as an independent OS?

W is for Windows in WSL, so no.

Based on reading this: https://github.com/microsoft/WSL2-Linux-Kernel/issues/1 I think the MS kernel would run fine without Windows.

If I were implementing it, I would definitely want to be able to test it using qemu without Windows.

What would be the point of a new linux distribution called windows server if none of your windows admin tools, scripts or binaries can run on it?

This is great, IMO. The sooner MS abandons Windows on Azure the better for everybody. You could argue that "monoculture" is not good, but, TBH, from a purely pragmatic standpoint, I struggle to name a single use case where I'd _not_ choose Linux over anything else when the use case is not desktop-oriented (cloud, edge, embedded).

I guess when you want to run software than only runs on Windows, such as IIS.

You won't find them here, but there are plenty of Windows-only IT service providers out there. It seems crazy that any system administrator, or programmer, in 2019 would only have any interest in Windows, but they exist and there are a lot of them out there.

seems to me that windows is still where you are going to find better (or even a driver) drivers for hardware. is this still true? if new cutting edge hardware released does it now have good linux drivers? (sorry i was scarred by the printing experience on linux)

Linux supports much more hardware than Windows. The situation with bad cutting edge hardware support applies only on desktop machines, where manufacturers don't bother with anything but the OS they put in the box.

Driver support is mostly a non-factor for cloud installs though. You can carefully chose what hardware you use, unlike the desktop, where the end-user can plug in any number of peripherals.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact