Hi,
We are considering a transition from OpenVZ to Xen for our web server infrastructure. The primary task performed by the virtual servers is to run a number of Drupal sites with a MySQL backend. The webserver and MySQL servers are on separate virtual hosts. We also have a number of smaller hosts running on the same hardware (a couple of dev-servers, a logging-and-monitoring server, a secondary webserver ++).
From what I have understood, there is very little performance degradation in
OpenVZ, also for IO operations, due to the use of isolated containers instead of actual virtualisation. I have also seen tests indicating detrimental IO performance of Xen IO [1]. Despite this, I have found several posters [2] suggesting the use of Xen over OpenVZ when it comes to running Drupal sites.
One of the main arguments against OpenVZ is unpredictable and variable performance because VPS-hosts tend to overbook their host capacitiy. This is not an issue for us since we host our own VPSs on our own hardware so we are in complete control.
So: Is there any reason Xen may be able to outperform OpenVZ for the webserver-setup I have described, or should I just continue to use OpenVZ?
Cheers, Einar
Sources: [1] http://www.ilsistemista.net/index.php/virtualization/1-virtual-machines-perf... [2] http://2bits.com/articles/hosting-virtualization-openvz-vs-xen-which-is-best...
On 06/08/2011 05:07 AM, Einar S. Idsø wrote:
Hi,
We are considering a transition from OpenVZ to Xen for our web server infrastructure. The primary task performed by the virtual servers is to run a number of Drupal sites with a MySQL backend. The webserver and MySQL servers are on separate virtual hosts. We also have a number of smaller hosts running on the same hardware (a couple of dev-servers, a logging-and-monitoring server, a secondary webserver ++).
From what I have understood, there is very little performance degradation in OpenVZ, also for IO operations, due to the use of isolated containers instead of actual virtualisation. I have also seen tests indicating detrimental IO performance of Xen IO [1]. Despite this, I have found several posters [2] suggesting the use of Xen over OpenVZ when it comes to running Drupal sites.
One of the main arguments against OpenVZ is unpredictable and variable performance because VPS-hosts tend to overbook their host capacitiy. This is not an issue for us since we host our own VPSs on our own hardware so we are in complete control.
So: Is there any reason Xen may be able to outperform OpenVZ for the webserver-setup I have described, or should I just continue to use OpenVZ?
Is your goal absolute best performance (why?), or simply adequate performance?
I wouldn't expect to see Xen outperform OpenVZ. However, if your CPU has has virtualization extensions, I would expect performance to be pretty close between the two.
If you're in need of better performance, I would look into tuning other traditional aspects. Find the bottleneck and alleviate it, same as if VMs were running on bare hw.
That being said, if you're considering Xen, I'd give strong consideration to KVM as well. It appears that the major players (RedHat, Ubuntu) are moving away from Xen toward KVM. http://searchenterpriselinux.techtarget.com/news/1511180/Linux-community-tur...
Cheers, Einar
Sources: [1] http://www.ilsistemista.net/index.php/virtualization/1-virtual-machines-perf... [2] http://2bits.com/articles/hosting-virtualization-openvz-vs-xen-which-is-best...
CentOS-virt mailing list CentOS-virt@centos.org http://lists.centos.org/mailman/listinfo/centos-virt
Hi Eric,
On Wed, Jun 8, 2011 at 4:33 PM, Eric Shubert ejs@shubes.net wrote:
Is your goal absolute best performance (why?), or simply adequate performance?
The hardware available to us is quite old, so pushing performance is absolutely the key issue. Stability/reliability and ease of maintenance are of course also factors. But since I've not have any bad experiences with OpenVZ in those respects, and I expect Xen to be at least as good, it pretty much comes down to performance.
I wouldn't expect to see Xen outperform OpenVZ. However, if your CPU has has virtualization extensions, I would expect performance to be pretty close between the two.
Unfortunately it doesn't. We're doing a "parallell shift" from a host with 2xOpteron 250 (2.4GHz) to another host with 8xOpteron 880 (also 2.4GHz but dual-core), so apart from the increase in number of cores from 2 to 16 and amount of RAM from 6 to 32GB there's no real difference to be found. But since we're transitioning anyway, we're giving alternative technology a consideration.
But okay, are you saying that Xen will at best be as good as OpenVZ using this hardware, and there's really not much point in me spending time learning Xen and running performance tests? That would save me quite a few workdays :)
That being said, if you're considering Xen, I'd give strong
consideration to KVM as well.
Thanks, I already looked at KVM, but it seems to require the virtualization bit, which means it is not an option for us.
Thanks for your input, Eric, it is much appreciated :)
Cheers, Einar
On 06/08/2011 09:13 AM, Einar S. Idsø wrote:
Hi Eric,
On Wed, Jun 8, 2011 at 4:33 PM, Eric Shubert <ejs@shubes.net mailto:ejs@shubes.net> wrote:
Is your goal absolute best performance (why?), or simply adequate performance?The hardware available to us is quite old, so pushing performance is absolutely the key issue. Stability/reliability and ease of maintenance are of course also factors. But since I've not have any bad experiences with OpenVZ in those respects, and I expect Xen to be at least as good, it pretty much comes down to performance.
I wouldn't expect to see Xen outperform OpenVZ. However, if your CPU has has virtualization extensions, I would expect performance to be pretty close between the two.Unfortunately it doesn't. We're doing a "parallell shift" from a host with 2xOpteron 250 (2.4GHz) to another host with 8xOpteron 880 (also 2.4GHz but dual-core), so apart from the increase in number of cores from 2 to 16 and amount of RAM from 6 to 32GB there's no real difference to be found. But since we're transitioning anyway, we're giving alternative technology a consideration.
But okay, are you saying that Xen will at best be as good as OpenVZ using this hardware,
This is my expectation. I don't know of any reason why Xen would perform better than OpenVZ in this situation. Someone may know better than I do about this though. I would hope they'd chime in here if they do.
and there's really not much point in me spending time learning Xen and running performance tests? That would save me quite a few workdays :)
I'd go this route, sticking with OpenVZ. Unless you've nothing better to do. ;)
That being said, if you're considering Xen, I'd give strong consideration to KVM as well.Thanks, I already looked at KVM, but it seems to require the virtualization bit, which means it is not an option for us.
If/when the time comes when you're migrating to hardware that supports virtualization, I'd give KVM strong consideration.
Thanks for your input, Eric, it is much appreciated :)
Sure. Just don't hang me if/when I'm wrong. ;)
-
Einar S. Idsø -
Eric Shubert