Survey: Proxmox VE Kernel with or without OpenVZ?

[kaptk2]

Keep the OpenVZ stuff. The performance is worth it alone. Proxmox VE initially attracted me because of the ability to do both OpenVZ and KVM. I am sure I am not alone in this.

If it goes KVM only there are other tools out their that do the same thing so Proxmox VE may get lost in the noise.


Proxmox is pretty unique in this reguard and I think by loesing that ability the project will take a major step backwards.

 

[tmo]

I agree with kaptk2, in that PVE is pretty unique in this regards and is what attracted us as well despite the fact that we mostly run kvms on PVE. Seeing OpenVZ support dropped would still be a shame in my opinion.

Martin, aside the obvious hardware and KVM feature support, do you think that KVM performance will increase either marginally or substantially using the newer kernel? And is the trade-off of loosing the OpenVZ support worthwhile at this point in time?

 

[Trulli]

We are not using kvm (as probably most professionals do).
No OpenVZ support => No proxmox for us.
Also we are bloody sure not to use a virtual machine to virtualize a virtual machine to virtualize.. ;-)

 

[typo3usa]

If you drop OpenVZ - then we have to drop ProxMox.
that's a lot of servers to change over :-/

I understand the dilemma however.

 

[Onyx]

We also use almost 100% OpenVZ. So if that get's dropped, we will have to look for an other solution.

 

[udo]

We use at this time app. 25% OpenVZ and i think this is a very nice feature. But till now we don't have trouble with the old kernel - all our proxmox-server runs well. Perhaps my point of view change, if we buy a new server and have problems to run proxmox on it.

I think the previous speaker are right, that the specialty of proxmox is the mix of kvm and openvz... but i would use proxmox furthermore also without openvz.

Udo

 

[martin]

It looks like OpenVZ plans to support the upcoming RHEL 6, but there is no official announcement yet.

RHEL6 could be based on Fedora 11 (also no official announcement yet) which means it will be a 2.6.29 Kernel - also good for KVM.

Release of RHEL6? Q1/2010?

So if all these rumors come true the community can expect the next development/maintained OpenVZ for 2.6.29 - but if you want to know it in detail, ask the OpenVZ team and I am sure as soon as they know it they will tell it.

... and of course, 2.6.29 will be a very good choice for Proxmox VE.

Best regards,
Martin

 

[Didelo]

I just would like to let you know that we also use Proxmox (currently on three servers) only because of the OpenVZ funtionality. No OpenVZ functionality would mean no Proxmox anymore, since we use Citrix Xenserver for our 'real' virtual servers.

Proxmox is by far the best tool to manage OpenVZ environments (and the only to manage mixed OpenVZ/KVM environments, if you are in it for that), but KVM/Proxmox for sure can't beat XenServer yet (in features and stability). Therefore I think dropping OpenVZ support would make Proxmox loose all its unique selling points, and I would strongly regret that!

 

[jmharris]

We have no use for OpenVZ, so using a standard kernel not only has the benefit of being up to date but makes it easier to perform custom kernel upgrades as we can ignore the OpenVZ patches.

We are considering this anyway- is it possible to roll a standard KVM compatible kernel without any OpenVZ and use proxmox with it?

 

[jmharris]

Interestingly in comparison to the other replies, I would say it appears as though virtualisation solutions like OpenVZ and Xen who continually have problems keeping up with kernel development have their days numbered.

RedHat are going with KVM in a big way. We have been using Xen extensively and are now looking to migrate to KVM in order to resolve stability issues (which other xen users have reported and they still haven't resolved..). At least KVM has got the backing of the kernel developers.

It seems that thinking is what Proxmox is going for here. I agree, although it appears they will have to maintain an older kernel for OpenVZ compatibility.

It comes down to are you going to handicap the KVM functionality because OpenVZ can't stay up to date with the latest kernel developments.

 

[holgerb]

This is really a tough question !

I´ve voted for newer kernel without OpenVZ since we are using KVM in the mean time. On the other hand I liked the support for OpenVZ. Esp. with some appliances available such as a Wiki. On the other hand (OpenVZ users may scream ) I´m a hands-on guy. So if I do not have OpenVZ as option I would simply use a Linux KVM on Proxmox to do the job. Performance penalties or not !

We are not using kvm (as probably most professionals do).

I disagree on this statement.

1st: Define "professionals"
2nd: Not all "professionals" (I strongly dislike this term) have the option to use OpenVZ. Our customers (for which we test our inhouse developed software) often use Win2k3 or Win2k8 as serverOS. So we have to run multiple Windows VMs for testing. A part of our customers are running Linux but as standalone machine but not a VZ container. So testing for their systems in a VZ container is not representative.

From a hoster / Linux server provider perspective things might look different though.

Proxmox VE is a pretty unique software.
I dislike VMWare because of being closed source and providing a restricted Linux with restricted command set and possibilities (be it the free versions or the commercial versions). I dislike XenServer from Citrix because of their licensing stuff. This is were Proxmox comes into game using KVM for virtualisation.

For me/us support for KVM and OpenVZ is a nice to have but not a must have.

 

[dietmar]

We are considering this anyway- is it possible to roll a standard KVM compatible kernel without any OpenVZ and use proxmox with it?

no - this is not possible.

 

[Didelo]

@ jmharris: I agree that Xen isn't that interesting and that KVM will replace it sooner or later. After all, in our testing, KVM beats Xen in speed, and with the support of Red Hat it certainly will gain stability as well.

I have no experience with Xen concerning stability. However, Xen is not exactly the same as XenServer, and XenServer beats KVM at the moment hands down on features and stability. Therefore I currently would never use Proxmox if it would only support KVM, since except for the speed of KVM I don't see anything where KVM/Proxmox currently scores better than XenServer (but maybe I'm missing something).

OpenVZ isn't comparable to either Xen, Xenserver or KVM. OpenVZ simply isn't a 'real' virtual server (since the containters use a lot of functionality of the Debian distribution they run on, while 'real' virtual servers more or less only use the hypervisor to gain access to the hardware) and has quite some issues with for example applications that reserve memory, but don't actually use it (like Java). But with the limitations OpenVZ has, it still has a unique selling point, and that is that it is super efficient if you just need loads of simple webservers or something like that. The memory management on the host is far more efficients than either Xen or KVM, which simply means you can create much more virtual machines than in Xen or KVM on the same hardware. Also being able to use older hardware without a virtualization-enabled processor is a big plus (in our case, Dell 2850's).

All in all, in our case we do have a use for Proxmox/OpenVZ in our production environment, alongside XenServer, but not for Proxmox/KVM. For sure KVM will gain importance in the future, but also be aware that RedHat will come with their own management suites, that might be quite interesting as well. Therefore, if I was the Proxmox team, I wouldn't put all my eggs in a KVM-only environment basket.

Just my 2 cents, of course... ;-)

 

[jmharris]

Didelo: Interesting points regarding OpenVZ, and I can certainly see why many find it useful at the moment. It seems to me it is a technical stop gap solution in most respects that will eventually die away, but right now it does have some particular advantages over KVM in resource utilisation and for non-vt/svm hardware.

This is not really the right thread for it, but how does XenServer differ from Xen? We have only been using the debian packages and compiling the Xen tarball releases. What are we missing out on?

I also agree with your comments on it being difficult to compete with RedHat if you are solely supporting KVM as they are likely to come out with some excellent support, but we prefer a web based system like Proxmox. Am I right in saying that RedHat only have an X client for config at the moment?

 

[gkovacs]

We chose Proxmox primarily as an excellent OpenVZ management interface, because our hosted solutions require the high performance and accessibility of kernel virtualization. There are no new kernel features that would compensate for the loss of OpenVZ.

Our KVM to OpenVZ ratio is about 1 to 10.

I think the combined support of Red Hat behind KVM and Parallels behind OpenVZ will soon bring us a more modern kernel. I sure hope they do, as ext4 support is badly needed for disk performance of VM servers.

In the meantime maybe we should test Proxmox with a newer development version of the OpenVZ kernel, like 2.6.26 or 2.6.27. Has the Proxmox staff contacted the OpenVZ developers? If not, maybe it's time for an official getting to know each other.

 

[tom]

We chose Proxmox primarily as an excellent OpenVZ management interface, because our hosted solutions require the high performance and accessibility of kernel virtualization. There are no new kernel features that would compensate for the loss of OpenVZ.

Our KVM to OpenVZ ratio is about 1 to 10.

I think the combined support of Red Hat behind KVM and Parallels behind OpenVZ will soon bring us a more modern kernel. I sure hope they do, as ext4 support is badly needed for disk performance of VM servers.

In the meantime maybe we should test Proxmox with a newer development version of the OpenVZ kernel, like 2.6.26 or 2.6.27. Has the Proxmox staff contacted the OpenVZ developers? If not, maybe it's time for an official getting to know each other.

The OpenVZ team knows what we are doing and we are in contact since the beginning of Proxmox VE. as you know, our vzdump is widly used also on non Proxmox VE OpenVZ installations, also in other OpenVZ control panels (new vzdump release is in the queue!).

If you take a deeper look into the history in the 2.6.26 and 2.6.27 development trees (see git on the openvz, also bugzilla) there is not that much activity and also 2.6.26/27 does not have more features than 2.6.24 - so this makes no sense to change for us. There are still OpenVZ features which are only available in 2.6.18.

Hopefully OpenVZ will publish a road map soon. This long thread shows that a lot of users really like container virtualization like OpenVZ but they are already unhappy with the unclear situation.

 

[Didelo]

@ jmharris: you could see XenServer as a (trimmed down) distribution centered around and optimized for Xen. Thus, you get 'guaranteed' and thoroughly tested stability and I think also more speed. As a very rough comparison for stability I think you could compare XenServer to RHEL, while Xen is more like Fedora.

Furthermore, you get all kinds of tools and even a terminal interface that isn't available for Xen. Last but not least, you get the XenCenter management tool that enables a lot of features for the less-technically oriented people; instead of having to use the command line, you just point and click and manage a cluster of servers and storage very easily. Of course, XenCenter doesn't have a web interface either, so I guess it isn't for you. But if you don't mind installing an application on your own PC, it sure works very well.

I recently haven't taken a look at the RedHat management tools for KVM. The last time I did, almost a year ago, it looked promising but still very early in its development stage. Either way, from what I remember, that indeed wasn't webbased either (I'm actually quite sure of that).

But let's get back on topic... ;-)

 

[samiux]

I am running Proxmox VE and mainly for KVM guests. I vote for KVM and latest kernel.

In my opinion, for those who are using OpenVZ is because of old hardwares. However, old hardwares consume more electricity and less powerful. It is not Green!

Newer hardwares are more powerful and consume lesser electricity in general. Newer CPU comes with more cores and you can virtualize more guests on a physical machine without performance compensation.

Once our old OpenVZed Proxmox VE machines died, we should buy a new hardware. If the kernel of Proxmox does not support the latest hardware, we will stop there and need to seek another solution. Then, the Proxmox seems useless at that moment.

Proxmox VE outperform others is her web based management interface. I think the development team may consider to provide two different kernels (or two different editions of Proxmox VE) for users to choose when install or download.

I use KVM on Proxmox VE as I can fully configure my servers easily without waiting for the release of the appliances. It is because I do not know how to make one for me.

 

[Trulli]

I disagree on this statement.

1st: Define "professionals"
So we have to run multiple Windows VMs for testing. A part of our customers are running Linux but as standalone machine but not a VZ container. So testing for their systems in a VZ container is not representative.

From a hoster / Linux server provider perspective things might look different though.

Professional: people running servers in a (even enterprise like) production environment and earning money with that. e.g. in datacenters ;-).


You use Proxmox/KVM for testing purposes.
Thats a completely different scenario...
And of course - depending on the software - you may want to test it in an environment that comes as close as possible to that of your customer. I can assure you that some things act differently in a VM than on real hardware (its getting better though ;-)).

Most will use Proxmox in a production environment where uptime etc. is mission critical.

KVM is at the present not yet ready to do that job (Fanboys are allowed to flame me now).
For Windows-VMs there may also be better ways and more stable solutions (and also free) out there right now.
For testing purposes even Hyper-V may be the better idea ;-)

Also, for everyone who uses "bleeding edge" technology when it comes to reliable and stable service delivery (e.g. systems not yet "industry proven / industry standard") i hope they have an angel hovering above them. Its russian roulette.

Proxmox is good because it supports OpenVZ and has a nice management interface - and offers the option to also use KVM, in case you need something else than Linux.
But again, for Windows-VMs there are better, even free solutions out there.

 

[andio]

There are bugs fixed between 2.6.24 and 2.6.26.

For me 2.6.24 does not work at all (IBM x-Series Servers x3850), so Proxmox with 2.6.24 does not work at all - while OpenVZ does work very well with 2.6.26.