Hacker News new | ask | show | jobs
by karavelov 1834 days ago
> PerformanceTest software only runs on Windows OS

I think this excludes huge population of servers as we know Windows is not dominant there.

2 comments

Are you sure? Statista has windows at a 71% server market share in 2019: https://www.statista.com/statistics/915085/global-server-sha...

Edit: Because people seem to think that the numbers are made up, here's some more sources with not quite as high, but similarly dominant numbers:

Red Hat: https://www.redhat.com/en/blog/red-hat-continues-lead-linux-... - microsoft @ 49% in 2017

t4: https://www.t4.ai/industry/server-operating-system-market-sh... - microsoft @ 47% in 2018

Are these real numbers or are they counting SMB instances running on commodity hardware and doing other weird junk to play with the stats?

Windows isn't even the most used OS on Azure. Certainly large deployments at Google, Facebook, Amazon, etc. accounting for a huge proportion of the population aren't running Windows either.

I don't have access to the source, but I literally cannot believe that is true.
There's often dozens of Windows servers installed on site in millions of offices around the world, doing various tasks (email, active directory, file storage, etc).

I'm sure if you looked at "servers installed in a data centre" then it would be closer to what you expect. But there are so many servers doing "boring" jobs in cupboards youll never see it totally skews the figures.

But if you looked at “servers installed in a data centre” what you’d discover is that there’s millions of servers in just one building.
Perhaps we could divide the statistics into "unnecessary servers Microsoft forced you to run" and "servers you actually need"?
This is so true! Most of the in-house data centers I had seen had Windows and they didn't really bother to even think about changing. It's totally different from cloud service providers, that some of them don't even provide a Windows server option.
it says it right in the description, which you don't need a subscription to read

> In 2019, the Windows operating system was used on 72.1 percent of servers worldwide, whilst the Linux operating system accounted for 13.6 percent of servers. Compared to 2018, both companies experienced an increase to their overall market share.

I don't think what is described as unbelievable is that the source claims that, but that the claim is true.

(And since it describes both “Windows” and “Linux” as “companies” that gained marketshare, I’m somewhat suspicious of their information gathering.)

But not being registered with them, I don’t have access to what passes for their detailed source information for that claim.

That's just unbelievable on the face of it.
I'm pretty sure any one of Amazon, Google or Facebook has more Linux servers than there are Windows server in the world.
Where are the numbers if you are so sure?
Google alone had 2.5 million servers in 2016 as per estimate [1], and 0.9 million in 2011 in [2].

So a reasonable guess for today seems > 5 million.

[1] https://en.wikipedia.org/wiki/Google_data_centers

[2] https://www.datacenterknowledge.com/archives/2011/08/01/repo...

I am missing the numbers for the rest of the world.
In the wild at actual data centers I'd gladly bet $10,000 it's actually under 1%.

It'd be the easiest $10,000 I'd ever make

You'd be losing that bet. Larger datacenters tend to use windows because of microsoft providing support contracts.

I don't have access to the full numbers, but I'd bet on the linux side being majority red hat for the same reason.

What planet do you live on? Have you ever been to a datacenter?

Look at top500, it's literally 0% windows, https://www.top500.org/statistics/list/

Many of the VPS providers don't even have windows options because the demand is so low it's not even worth offering as a product.

Server farm sysadmins I know habitually lament they are rusty in windows whenever it's in front of them because it's completely vanished from their daily experience

If this was a survey of companies where the weight of say, a liquor store with a 20 year old beater in the back closet that they'll call a server is 1 vote, and say, all of Amazon is also 1 vote, then maybe you can get that number, but that's basically the only way

top500 is a super computer dataset, it's not really relevant to general server market share

you could even look at Red Hat, one of microsoft's competitors, who showed microsoft's market share at 49.6% in 2017: https://www.redhat.com/en/blog/red-hat-continues-lead-linux-...

And over on wikipedia it says 24. https://en.m.wikipedia.org/wiki/Usage_share_of_operating_sys...

So the position moved 71 -> 49 -> 24

There's a difference again. Is "server" a chassis in a rack with redundant power and some RAID array or is it some standard computer that was christened as such? It's not a useful category because it allows for arbitrary segmenting to pump the numbers. Are we counting by service, machine, company? Depends what result you want.

If you're going to count windows machines (as in cases) by hand at a Noc, farm or center you won't be needing a calculator or likely a second hand.

Windows is certainly more prominent than Solaris or any commercial traditional unix but outside of desktops, which have a cleaner definition, it's been going in same direction for 20 years.

Microsoft has also been pivoting away from windows as a revenue source for a while. Video consoles, media companies, Azure, LinkedIn, GitHub, they don't necessarily see windows in their future and we shouldn't either

Not all servers are web facing or live in data centres. There are tons of small businesses that run servers in a backroom. Those are predominantly Windows.
right but that analysis isn't useful. If I used the same logic to determine a solution for hauling a large load by vehicle and looked at how most cubic meters of things are hauled, I'd probably be forced to conclude that a sedan trunk is the best option. However that's only due to the preponderance of sedans.

It's a very careful way to state the question to make it look like the numbers back what's clearly a wrong answer.

It puts the framing of "well I guess I'm using a sedan" and resituates the problem to solve all its shortcomings. It's a classic implicit framing propaganda technique.

I work for a hosting provider. Windows exists and it's used for corporate infrastructure, but the vast majority of actual applications and support services are running on Red Hat, CentOS or Ubuntu.

Windows trends to be reserved for AD, jump hosts (much to my chagrin, I prefer SSH), virtual desktops and stuff that end-users access, because user management is where Windows is still ahead.

I suppose in small businesses they will have all the user management bits but no need for more than one or two "real" servers, so it doesn't matter if it's Windows too.

In bulk, Linux servers are much easier to manage though and since most application platforms will work just fine if not better on Linux, it's just the sensible choice unless you have something specific that requires Windows. Generally easier licensing certainly does not hurt.

The fact that on any Linux distribution you get trivial, easily-managed access to thousands of software packages free of licensing worries out-of-the-box makes for a large advantage over Windows where the concept of using a package manager and central repository seems to be a fairly new thing and you have to set everything up yourself or just deal with developers and users installing random crap from who knows where because by default there's no better option. There's also the fact that Linux server software tends not to have GUIs because servers are headless by default, which also makes them easier to manage once you run out of fingers to count them.

Why do you need paid support for your OS?
It's a requirement still.

Plenty of open source people have offered paid support so corporate clients could expense things.

Not only does Microsoft offer comprehensive support packages for their products, but in fact the TCO of a Windows Server system is almost always lower than a Linix system while also having the better user productivity. Considering this, any business that is serious about digital transformation would do well to indeed reach for Microsoft’s portfolio.

Of course it‘s easy to make up statistics to show that the OS made by some Sovjet hacker in his mother’s basement (probably using lots of stolen MS IP) is supposedly more popular than the products made in the US by a reputable firm such as Microsoft, but who is going to believe that?

I used to have a small startup with over 50 servers per engineer. Tell me all about TCO and digital transformation, I really need to learn about that stuff.
I wonder if this only considers the OS that a server was shipped with? Which is typically nuked as soon as the customer receives it.
Servers aren't generally delivered with an unwanted OS, or an unwanted Windows license.

If it's going to run Linux, you tick that option when ordering. Or you tick "no OS".

This may be true for small companies, but larger businesses tend to use windows due to support contracts.
Larger companies have no problems finding people to pay to support Linux, or if they are really large (say, Amazon), supporting it themselves and being who people (including big companies) pay to support and host Linux systems.
it all depends on size. some larger (typically older) companies with smaller engineering teams will make the tradeoff you are describing. But the big players today, who manage hundreds of thousands to millions of servers are not going to pay per machine / per core license counts for support. They build and maintain most of their own software for their own needs.
How many people outside of a homelab/lab going to run a performance test suite on a fleet of servers? The sample size must be interestingly skewed.

This may be more useful information for the server market, even if old.

https://www.tomshardware.com/news/amd-vs-intel-q3-2020-cpu-m...

It’s pretty common actually. A server buy can easily be an 8 figure deal, getting servers that turn out to be stupidly configured for your apps that your stuck with for 5 years is exceptionally dumb.
I should've emphasized lab before of homelab. Agree, doing a full stack hardware validation (managed to even get a full multi-vendor stack deployed into a Dell validation colo to do our internal performance test). Even then, we only ran out own performance tests, as we cared about our workload, not a generic workload suite.

Edit: Additionally, we only did that work on a single set of systems prior to purchase, not the systems post purchase, so 1 of N, instead of rerunning on N. It was presumed that performance would remain the same on same hardware configuration etc.

That's a telling edit. It's easy to misconfigure a few machines, or have an OS update hurt performance of your entire fleet.
When I had a fleet of servers, I ran a performance test suite after every boot to make sure the machine was OK.