Survey: Proxmox VE Kernel with or without OpenVZ?

[alain]

That may very well be the case, but being able to SUPPORT kvm, and actually RUNNING kvm are two totally different things.

In my experience, you don't need a very powerful machine to run KVM (or other virtualization solutions). The main requirements are memory (very cheap now) and storage.

And although the demise of some older machines is in sight, they are not going to all suddenly disappear mid next year. They will be around for a bit still to come. (IE6, anyone?)

Yes, but you can use the current releases of Promox for these older machines.

The main point is that openvz does not seem to evoluate very fast in the last times, but KVM is developping very fast.
So it makes sense to have a separate branch to follow the KVM pace.

For example in proxmox VE 1.4b, the new storage model is only available for KVM :
"Currently only KVM guests can benefit from these enhancements, containers still need to be stored on local storage"

Alain

 

[andrea]

About storage for containers, it should be possible to mount external storage on /var/lib/vz/private, assuming the external storage is available to the HN, regardless the storage model used.

Andrea

 

[Omny_Devi]

In my experience, you don't need a very powerful machine to run KVM (or other virtualization solutions). The main requirements are memory (very cheap now) and storage.

The main point is that openvz does not seem to evoluate very fast in the last times, but KVM is developping very fast.
So it makes sense to have a separate branch to follow the KVM pace.
Alain

I would say those are a few of the "main" requirements. You need a bit more than a lot of RAM and a few Hard Drives.

KVM can evolve as fast as it likes, I greatly doubt it will evolve so fast I would think that throwing out VZ would be a good idea. They just work too well together side by side for me to bite on that.

 

[aneubau]

We are using both OpenVZ for Linux machines and kvm for Windows servers in production use. We use 4 physical server and have separated kvm and openvz virtual machines. So we would prefer to have two kernels at the moment as we could run two servers for kvm and openvz machines. I know not everone is in this sitiuation but could it not be a solution to keep the current kernel and have a second one optimized for kvm only ?
I also feel that kvm cannot be a replacement for openvz at the moment but did anyone do a real comparison between both virtualization technics (with both optimized kernels) to see how far the performance differences are ?
At the moment it seems that openvz hinders optimal usage of kvm without having advantages for openvz and this is not the way it should go.
Another question to this:
Would be Linux Vserver an alternative ?
I have used it befor I changed to Proxmox in combination with vmware server and it worked flawless.

 

[plewka]

I'm using ProxmoxVE in Workgroup for very different guests. Only some of them are (differrent) Linuxes...full virtualisation use.

At home I'd like to virtualize, too. But I need support for IOMMU and moving PCI-Hardware into the guests (LinVDR and others).
Because of this I'm in need for that possibility which started in Kernel 2.6.28.

 

[lioenrio]

i have a 200 guest in openvz
and 15 with kvm

You know if virtuozzo suport new kernels?
Its posible move guests to vituozzo containers?

Thanks

 

[mangoo]

2.6.32 kernel will support KSM - Kernel Samepage Merging feature.

For example, when running lots of similar guests, memory pages with identical contents will be "merged", thus saving the memory.

This feature can possibly make KVM use even less memory than OpenVZ (although some tests would be needed here).

 

[dietmar]

This feature can possibly make KVM use even less memory than OpenVZ (although some tests would be needed here).

OpenVZ will also gain from KSM.

 

[mangoo]

OpenVZ will also gain from KSM.

Not really, for the following reasons:

1) OpenVZ would have to support 2.6.32 or newer kernel - long term, fixable, but rather not in foreseeable future,

2) application using KSM has to support "madvise" (MADV_MERGEABLE, MADV_UNMERGEABLE...) system call. Here, kvm is one process and supports this system call; it doesn't matter what processes does the kvm/guest run.

With OpenVZ however, all guest processes are visible for the host's kernel. So to support KSM with OpenVZ, you would have to add "madvise" system call to all processes running in OpenVZ container.
I don't see it very likely that someone will add "madvise" system call to bash, perl, python, java, httpd, php, MySQL, PostgreSQL, X, firefox, and all sorts of random foo&bar programs people may be running in their guests.

 

[dietmar]

2) application using KSM has to support "madvise" (MADV_MERGEABLE, MADV_UNMERGEABLE...) system call.

I am sure there is a workaround.

 

[mangoo]

I am sure there is a workaround.

I haven't heard of any.

I can imagine one (which doesn't require the binaries to be modified) using existing kernel infrastructure though: use in-kernel binfmt_misc to execute every binary in the guest with a special "interpreter".

It would use additional system resources though and would need installing additional software in guests.
And would likely fail for some more complicated programs.

Also, such software does not exist.

Unless you've heard of other workarounds, or have different ideas.

 

[user100]

1) OpenVZ would have to support 2.6.32 or newer kernel - long term, fixable, but rather not in foreseeable future,

Good point! If the OpenVZ guys would support a newer kernel in near future then this thread can be closed, right? Currently 2.6.18 is still the stable OpenVZ kernel. So I think it´s clear why this vote was started? When "new-features AND openvz" was an option this vote would not really make sense?

Greetings,
user100

 

[dietmar]

I haven't heard of any.

If I understand them correctly, 'madvice' is just an optimization (do not use all memory for KSM). But you can also scan all container processes - or why do you think that is not possible? Sure, maybe you need to add another syscall.

But please let us use the mailing list to disuss such technical details.

 

[donty]

My vote is for a dual approach, OpenVZ+KVM old kernel and KVM new kernel so that if you do use KVM or OpenVZ you are not penalised.

OVZ is good but doesnt isolate the VMs completely, it rests on the underlying shared kernel not getting blasted. If it does they all fall down. You also don't have the kernel flexibility and there are some gotchas that I ran into with OpenVZ. I know you can accommodate most of them with fiddles and that is great if it is a one time issue replicated over 100s of servers but in our case it isn't what I think of as clean and simple.

We dont virtualize just to squeeze the last drop from hardware, it is the flexibility and control that we like, so as usual it is horses for courses, that's why I think that neither should be shot in the foot - OpenVZ has good uses, as does KVM. We use both in different ways.

Whatever happens, thanks for the great work on Proxmox!

 

[hzqbbc]

Hi all!

Proxmox VE uses currently a 2.6.24 based Kernel. Due to the limitations of OpenVZ there is no actual Kernel possible (OpenVZ 2.6.26/27 are quite similar and also quite old).

So the question is, should we go for the latest Kernel to get the latest and greatest KVM functionality and best hardware support?

What do you think? Please vote!

If openvz is removed from PVE, i guess over 60%+ end user will abort PVE, because there are more powerful kvm management tools will come out, eg: ovirt is cool.

For my part, we choose PVE due to perfect KVM and VZ integration, PVE is unique, so please keep openvz, it will attract more users

 

[littletiger]

A quick observation on kernel versions, which I hope the gods around here might be able to shed some light on:

Red Hat with its massive investment in KVM (they bought the company) is still using 2.6.18 kernel in its recent Red Hat 5.4 release.

Yet we are saying that the 2.6.24 kernel (which Proxmox is on) is too old because KVM is moving ahead so quickly.

How is Red Hat handling KVM with their 2.6.18 kernel ?

Do they only support cool stuff like PCI Pass-Through on Xen ?

 

[tesquenure]

Im just a small user.

I try Vmware and Ubuntu KVM berofe proxmox.

I love PVE cause a baremetal as Vmware but with KVM.

KVM hes much more heasy for me who comme from windows.
Install OS on virtual PC dont change my knowlage.
Its juste like a new motherboard, this some new switch.

I use it for my small office.
Before i have 6 PC as small serveur.
1 IPCOP
1 SME 7.4 Administrative software
1 SME 7.4 Fax and mail server
1 SME 7.4 OCS, GLPI and WIKI
1 SME 8 beta for test
1 Win XP Pro with special configuration use it from lan and wan (i use Ubuntu on my laptop, some time i need Windows).

At this time i pass all my SME on 1 PVE, my windows too.
And try to use PfSense on my PVE

My second PVE use for test and as backup.

For shure Proxmox hes a green IT ^^

Small user as me juste want KVM, i try openvz wiki, its takes me more time to use it and upgrade than install turnkey wiki.(because i dont realy know how its work, and not because KVM is better than openvz in this case).

For a linux JEDI hacker, im sure openvz can be better.
But a linux JEDI hacker, use is own system

In a other way, who pay service to proxmox ?
Small user as me (no at this time) or peaple who provide internet service(and they use for shure openvz).

All user in my case can migrate on a new solution with KVM if they find a new mature one.(more turn office use as USB, PCI card and new video card vith virtualization), but
openvz user with small windows internet providing, realy need Promox as its.

ok ok i go to sleep skype makes me crazy.

Regards,
Tesquenure

 

[charon]

i think it would be nice to choose during installation.
there are servers that need openVZ and other that only need KVM.
The proxmox Server without openVZ support could always run on the bleeding edge KVM version.

 

[lwilke]

I wonder how lacking a proxmox version with lxc and kvm would be in comparison to, say the current 1.4?

 

[dietmar]

I wonder how lacking a proxmox version with lxc and kvm would be in comparison to, say the current 1.4?

Our current lxc prototype is not usable in a production environment, sorry.