I was working in a lab to validate if you can successfully use Xen on CentOS 5.4.
Trust me, I really tried.
- Updated to a 5.5 kernel
- updated Xen and libs to 5.5ish state
- Installed pypxeboot and udhcpc to manually wrange with PV PXE support
- Messed around with libvirt’s idea of network config, multiple bridges and / or vlans are much less fun if you have to maintain XML files for them.
- found a 2007ish bug report because the HVM-Linux IO performance on i386 arch was around 200KB/s
- Changed dhcp config because I didn’t find a udhcpc version with the patch to support arbitrary ID strings (so you can be “PXEClient” instead of “udhcp 0.48”
- Hacked our disk partitioner
At least I couldn’t even create PV domUs, they crashed during network setup. Commenting the whole “vif” line did not help.
I thought I really had to reboot the system. During the reboot it just happened that I somehow ended up in the Raid HBA config, created a sexy raid10 array and by mistake inserted an Oracle VM CD.
At times I even might have pressed “OK” “Install” “Remove all partitions”…
Let me see if I can have the server & VMs up by tonight.
In the end, it was good to do the testing, but the result is without question… with a RHEL/CentOS you get like 50% of Xens feature set, too little bug fixes and, alltogether, just the worst of all worlds. Oracle VM i.e. comes with pypxeboot out of the box (not saying that it will just work… but still…), and tmem is also available.
Alternatively I’ll go with Ubuntu 10.10 or OEL6 and enjoy a more current Xen 4.1, but that would put me back at “intregation? that’s what the end users do!”. 🙂