Red Hat Wants Xen In Linux Kernel

DIY News writes "Red Hat is aggressively pushing to get Xen virtualization technology included in the Linux kernel as quickly as possible. This move comes as Microsoft is pushing its own virtualization products and recently relaxed some of its licensing requirements around Windows Server 2003 to facilitate more pervasive adoption and use of those technologies."

Comments

Xen into kernel

[b100dian]

What exactly does "virtualization technology included in the Linux kernel" means?
That you can run virtual machines with that kernel? that that kernel can be hosted into a virtual machine?

Or that you can install parallel kernels and run part of the ELF binaries on the other machine?..

Re:Xen into kernel

[caseih]

Since you've received few answers to you actual question, getting xen into the kernel means the xen patches required to run the linux kernel in a xen hypervisor (both as a guest and a host) will be a part of the mainstream kernel and be able to be built trivially. RedHat ships 3 different kernels now with FC4. One is a normal kernel available in both smp and non-smp configurations. Then we have the XenU kernel, which is a kernel designed to boot in a guest xen session. The Xen0 kernel is the kernel that you'd actually boot on top of xen and use as your main OS.

Once the Xen0 kernel is running on top of xen, you have basically a normal linux kernel running that does all the hardware support. Then you load up Xen guest machines running the Xen0 kernel and these run their in their own virtual machines complete with their own disk images and linux distro. So xen doesn't really have anything to do with running elf binaries on the other machine. If you ran FreeBSD in the guest, it would run those binaries inside of that OS and that libc. When Xen 3.0 comes out, if you have the new intel or amd chips that support on-chip virtualization, then Windows XP can even run as a guest underneath the linux kernel-Xen0 host.

Re:Xen into kernel

[secureboot]

Don't let that comment fool you though. Xen is much, much more. What if you organization had 4 distinct sites they wanted to host on one server? Start up 4 virtual machines, and back up their running state from time to time. If one goes down, just restart it from your clean backup IN SECONDS. Better yet, do it automatically. At the same time, Xen enforces separation from the host OS that the virtual machines are running on, so you don't have to worry about it being compromised (well, not in any way anyone has been able to demonstrate or even postulate yet).

Re:Xen into kernel

[hey!]

easier management of computing resource, unless I miss my mark.

In rough terms:

Admin Cost = N * (H + S)
where N is the number of computers, H is the network and system hardware admin costs for a single machine, and S is the sys admin costs for the machine. Distributing:

Cost = NH + NS

This is a gross simplification, since we all know that complexity is not a linear function of network size, but it will do to be goign on with. Now we take NH and by virtualizing the machine it becomes simply H, so

Admin Cost (multiple virtual machines) = H + NS

Basically, I think it'll be common practice in the future to create virtual machines out of thin air by copying a config file or some directories on a machine with available bandwidth. If the cost of enough surplus hardware is less than (N-1)H, then wouldn't it be cheaper to virtualize?

Of course the complexity is that costs aren't linearly related to N, or for that matter constant in the size and class of machine you are managing. Which is another way of saying YMMV. I think there's clear application in many kinds of situations, for example in software development where we're constantly worried about the various combinations of software our work will have to coexist with. It'd be very convenient to be able to pull a certain system configuration out of a library and have it up in a few minutes, then trash it after a few hours of use. But it may have potential in production environments too.

Xen...

[Anonymous Coward]

Xen isn't all that hard, you just need some jump boots and a particle weapon of some sort.

Umm

[interiot]

Well, Xen is free, and Intel/AMD hardware solutions are comming soon, which will allow Xen to run Windows unmodified. So, once everyone is upgraded to the new CPU's, virtualization will become a basic standard feature for everyone. MS can compete by giving their solution away for free, but either way, it doesn't get better than free for the consumer.

Re:Umm

['nother poster]

Microsoft got in trouble because the integrated their browser and media player into the operating system. If they would simply have had a browser and media player on your Windows install CD that you could install, or not, there wouldn't have been nearly the ruckus. What they did was make the install of the items required, and made it impossible for the average user to remove them without crippling the operating enviroment.

Operating system virtualization on the other hand pretty much requires it be hooked into th OS by its very nature.

Two birds, one stone

[bernywork]

Not only do they get the ability to knock the shit out of Microsoft, by taking away the base platform from them, they also get to try to get some market share from VMWare.

Imagine if you would the ability to use Xen for unlimited operating systems, no licensing cost of the base OS, thinking about it, I would prefer to be in Microsoft's shoes as opposed to VMWare's. Only difference is that Xen when compared to VMWare is a very immature platform and no IT manager is going to take Xen over VMWare just yet (Unless cost is a BIG factor).

I would have to say that this is still very cool, with all the new Virtualization options come out in the new cores shortly and if they can get to market before Microsoft, this is a great way to pick up some customers. Kudos to RedHat and IBM and Intel and everyone else for making this happen.

Aggressively pushing?

[gringer]

"My goal is to get this done in the most collaborative way possible with anyone in the community who wants to participate," Stevens said, adding that Red Hat is committed to putting on this project enough of its staff who have the technical knowledge necessary to get the work done.

Perhaps it's only me, but this doesn't sound aggressive; this sounds friendly and cooperative.

(c) me, 2005

[Library Spoff]

*hmmm*
Must remember to patent the idea of trojan/virus that uses visualisation to run a spam/DOS server
in a Windows environment...

Rubs hands with glee as he tries to sell the idea on IRC.

Re:It means

[Octorian]

I still remeber reading that the whole x86 architecture didn't meet the requirements for virtualization, meaning that this recent trend is probably the result of VMware figuring out some "tricks to make it work", and then everyone else jumping on the bandwagon.

In any case, if you really want to learn about the fundamental concepts behind virtualization, I strongly recommend reading the following paper: Formal Requirements for Virtualizable Third Generation Architectures

Yes, it was published in 1974, but most of the concepts are still very applicable and make a lot of sense. (though the architecture examples are obviously dated)

This is a very good paper which lays out all the ground rules. Sure, it may sound a bit academic in terminology and explanation, but it is still quite readable.

Re:How does virtualization work?

[Waffle Iron]

But then what do you when two or more OS instances want to monkey around with hardware that has state?

The Schrodinger Corp. makes special PC cases that can handle those requirements.

It's not personal, it's just marketing

[FishandChips]

The Slashdot summary is a bit misleading. What the article says is that Andrew Morton has been expecting a kernel submission for Xen for quite some time now but a) has yet to receive it, and b) needs to go through the usual process with other "stakeholders" before any incorporation. Later the article quotes the Xen folks themselves who point out that "feature creep" and the need to generally get things really solid and stable has made everything take a little longer.

What the article actually seems to be saying - it uses the word "agressive" a lot as if this was some kind of virtue - is that Red Hat has a new senior honcho who'd like to make his mark. The issue of incorporating virtualization technologies into the Linux kernel is taken as a given by all parties. Which is hardly news. Chalk one up to the Red Hat marketing department for a nicely planted "news" story about their increased investment in the area (new hires, etc.), perhaps.

Re:What?

[LinuxHam]

Multiple OSes at once? [..] I never got the whole concept of virtual servers

Its mainly an enterprise play.

If you're an old-timer UNIX admin, you may have difficulty understanding the point of server virtualization (i.e. multiple OS instances). In UNIXland, it has been normal and customary for several completely unrelated applications to run under the same OS instance, together servicing thousands of users. That never worked well in Windowsland. That being said, it didn't stop manufacturers from making staggering improvements in performace and capabilities of Intel servers. Companies grew to expect single Intel boxes to perform at the levels of large UNIX servers. The only way to achieve that in recent years has been to use industrial strength virtualization technology (basically, ESX). The boxes are beefier than ever, and ESX isn't cheap, but it works wonders.

For $50k, you can run 50 VMware guests on one very beefy box (not counting SAN), but you'd want a second for failover. For $75k, you can run about 100 guests on 14 blades in 7U (again, not counting SAN) and have the guests automatically migrate to the blades most able to run their workload at that moment in time. Ask a blade to come down for maintenance, and all the guests scatter to other blades before the blade powers off. Replaced a dead blade with a blank? Your systems management policies detect the new blank and automatically install ESX on it so guests can migrate back and evenly spread out the load.

Sounds crazy, I know, but that's a taste of what we're doing in the enteprise space these days.

Re:It means

[Chirs]

The full x86 architecture is not suitable for virtualization, because there are a few instructions which fail silently when run from user level.

VMware uses various techniques to get around this, including full simulation and binary re-writing.

Xen uses another approach, where they port to an instruction set that is basically x86 without the problematic instructions. This approach requires that the guest OS's be modfied.

This will all change with the new virtualization instructions being added by both AMD and Intel. Once that is in place, Xen will be able to run unmodified guest OS's (such as Windows, for instance). There will be a speed hit though, so modified guests will be prefered if speed is an issue.