I’m working on doing some tests to compare different virtualisation technologies. Whilst those tests are far from finished, I got some numbers this afternoon who give a quick first impression.

What is being tested here?

• Running some specific processes
• Timing how long they run
• On a virtual guest (1 for now) on different platforms

Host environment?
Five servers, each running a different virtualisation platform, running on a machine with 2 quad cores Intel(R) Xeon(R) CPU E5520 @ 2.27GHz with 12GB RAM. One 160GB SATA 2.25″ 5400RPM disk per server, of which around 120GB is used for the guest disks (LVM).

The 5 different platforms are:

1. VMWare ESXi 4.0
2. Red Hat / CentOS 5.4 with Xen 3.0
3. Red Hat / CentOS 5.4 with KVM 83
4. Debian Lenny with Xen 3.2
5. Ubuntu Karmic with KVM 0.11

Guest environment?
All the guests are running Debian Lenny, in a vm with 512M Ram, with 2 virtual processors. And a file system of around 6GB. Virtual disks are LVM logical volumes on disk on the server host. Xen guest are runnig in paravirtualisation mode with the domU kernel available in Lenny, KVM guests are configured with virtio hardware. VMWare emulates Intel(R) PRO/1000.

There are 20 clients deployed on each server. So far I ran tests on 1 concurrent guest only. The dual concurrent client tests are running whils I write this, I’ll hope the scripts will keep running during the weekend

A quick peek in the test logs showed me following numbers. The items should speak for themselves, tthey are noted as

platform-test=time_in_seconds_to_run

A smaller number means better performance.

Lame
Converting 10 wave files to mp3

debxen-lame=44
ubukvm-lame=45
vmware-lame=45
rhtkvm-lame=47
rhtxen-lame=53

Unzipping a kernel tarball
debxen-zip=78
vmware-zip=78
ubukvm-zip=81
rhtkvm-zip=83
rhtxen-zip=94

Untarring the tar archive
vmware-untar=25
debxen-untar=27
rhtxen-untar=27
ubukvm-untar=30
rhtkvm-untar=40

Compiling that kernel
ubukvm-kernelcompile=3789
vmware-kernelcompile=3879
rhtkvm-kernelcompile=3918
debxen-kernelcompile=3999
rhtxen-kernelcompile=4906

The biggest thing to note would be a lesser performance of Redhat + Xen when it comes to processor load, but keep in mind that this is an older version of Xen. On the other hand The respective older version of KVM on Red hat plays rhather well. I expect Xen to perform better when it comes to disk access.

There are other tests being processed also (iperf, iozone, tbench, dbench) but I can’t give a quick overview of those as the time to run is irrelevant for those.

More to come.

Being a proprietary tool, VMWare has a problem in common with Windows: it’s full off proprietary API’s you have to go through to be able to pull a backup. Also, whilst VMWare (ESX, but in practice also ESXi) has a unix/Linux like management environment, you don’t get all the tools you’d want to do your thing. Rsync to name one.

Of course, I you walk the line, buy sufficient licenses and extra software, you get to use VCB and can interconnect with “Supported” software like Symantec Backup Exec and the like.

But here and know, all I wanted, next to regular backups, was being able to easily pull backups to a Linux backup server (just a host with plent of disk space, and all the standard software I wanted.) Now, I found it hard to find the right solution to do that. Lots of custom scripts, but nothing that quickly worked. Often ugly scripts or hacks.

So far, the best solution i sticked to is the GhettoVCB script. Add a an NFS export from the Linux backup server to VMWare as a Datastore, and after a simple config, just run the script.

Downside of this approach, when used for a full ESX with SAN environment, is that you have to run in on the virtual host machines themselves, you can’t grab it over a separate VCB host which can haz its own SAN connection.

It works, but I still feel it’s awkward, not sure why. Anybody other suggestions for plain and simple backup tools for vmware?