LinkedIn Migrates Their Servers From CentOS To Azure Linux

[rmfx]

Microsoft not using Windows Server on their servers? Whoa.

Nobody wants to eat their own excrement.
They're lucky the world is packed with flies.

[Ironmask]

Microsoft not using Windows Server on their servers? Whoa.

Probably for a couple reasons:
1. The majority of LinkedIn's infrastructure was probably built on Linux and probably either only works on Linux or works better on Linux, and sticking it in a VM (WSL 2) wasn't a compromise they were willing to make.
2. The team probably has no experience, training or certification for Windows Server, and training them would be a nice waste of money, which also invalidates any reason to try to get their infrastructure working on WSL 1 as well.
3. MS doesn't care about their software anymore, they're a cloud provider now, and they don't care what you use as long as you use their servers. Forcing people to use something else would actually be a determent to their business model. Azure Linux itself is probably a direct result of this outlook.

[cjcox]

Red Hat: We need to end the whole community clone RHEL thing.

World: Do you think that will cause any problems?

Red Hat: No.

[timrichardson]

This is a milestone moment for Microsoft: LinkedIn was at the time Microsoft's largest acquisition and it remains the second largest. This is a sign that Azure and Microsoft's linux support on azure is ready for anyone, I would say. Clearly a lot of work has gone into it. The cost of migrating to Windows would have been ridiculous and I doubt it had ever a moment of serious consideration. Also, it probably would have failed anyway, which would be humiliating. And to what benefit? I doubt it would change sentiment about Windows Server. It's a legacy technology and it will be around for a long time. But a lot of the investment in the Windows tech stack seems to be better interop with Linux, or reimplementing Linux technologies. Search for "windows server innovations since 2010" is pretty sparse reading. Yes, you can run kubernetes for all your windows container requirements (has anyone ever seen a windows container?), and windows server has a headless mode now. And there is COW filesytem too, somewhere or other.

[danger]

There is nothing interesting about Azure Linux. It's still CentOS-based. Packages are rpm. Package manager is dnf. The structure closely resembles CentOS, only with much fewer packages. You can check available packages here: https://packages.microsoft.com/azure...3.0/prod/base/

[pWe00Iri3e7Z9lHOX2Qx]

There is nothing interesting about Azure Linux. It's still CentOS-based. Packages are rpm. Package manager is dnf. The structure closely resembles CentOS, only with much fewer packages. You can check available packages here: https://packages.microsoft.com/azure...3.0/prod/base/

Except that isn't accurate at all.

A Build attendee asked what distro Azure Linux was based on. “Azure Linux is its own distribution. We didn’t fork Fedora or anything like that. We have borrowed code from them, it’s an RPM-based distro,” said Perrin. “The reason we chose not to fork a different distro … Microsoft has kind of a history with Linux. … I think the Balmer quote is from 2001, but a lot of the sentiment still lingers even today. Part of the reason we did not choose to start with a distribution and then fork it for our needs is we didn’t want to be seen as doing the embrace and extend thing again, we didn’t want to wake any of that up, we figured, build it from scratch, we can tailor it to our needs … we’re scratching an itch we had and offering the solution back to the community.”

SUSE (and other distros) use RPM and you can even optionally use dnf there. That doesn't mean they are based on CentOS.

[toves]

I am curious what for Linkedin would be the need software raid on Azure.

The virtual disk devices their VMs would be using would presumably already be redundant and snapshot and archived.

The blog reference conjoins the reference to XFS with the reference to software raid.
AFAIK XFS on Linux doesn't have any raid capability - I vaguely recall XFS on Irix 6.5 might have had some volume management capability (probably only projected and mentioned in the SGI man page.)

So I assume the raid is from LVM (lvcreate) or MD (mdadm.)

[Mavman]

Anyone else seeing the M$' EEE strategy here?

[Ironmask]

Anyone else seeing the M$' EEE strategy here?

I don't think dogfooding counts as EEE.

[darkonix]

Anyone else seeing the M$' EEE strategy here?

Ah, now that you mention it, no.