[SOLVED] Best practices, running services on the host, QEMU

Hi all,

I remember reading in the OpenVZ manual that it was against best practices to run anything apart from OpenVZ on the host. Is this the same for QEMU? I cant find it in their docs.

The reason I'm asking is that I need a block level passthrough so I can use the VM as a nfs/glusterfs server... could I run a iSCSI target on the host and an initiator on the guest (I know that itself is against best practices, but I have read up on it and I think it will be fine for my purposes). Or I could just run glusterfs on the host... but that just seems to easy.

I'm trying to integrate the storage with the processing nodes in a cluster while still maintaining HA, its not going well so far!

Thanks for any input.

Here's a nice classification of different virtualization solutions: http://en.wikipedia.org/wiki/Platform_virtualization. As for doing your actual question I'd say just go do it: it's not like it takes ages configuring it and if it works you can tell us about your triumph.

Any best practice on one VM tends to be similar to each. Best practices are for security, efficiency and data integrity.

As a general rule, any server should be limited to only the essentials needed to perform it's work. Extra apps and exposure can lead to less than optimum performance and usefulness.

In a very strict sense, a VM doesn't improve security, efficiency. It's usefulness is more for utilization of resources, ability to clone and replicate and maybe a few other uses.

So I now have a setup that uses "multifunctional" nodes for storage and processing using qemu libvirt and glusterfs and as instructed I'm telling you all about my triumph..!

As requested below, I will write some more stuff here when I have a sec

Since you have this post, why not go ahead an post what you have?

A few months back I decided to be extremely clever and did this:

1)
On my host compiled its Kernel with all host and client virtualization options (KVM) turned on .

2)
After having compiled the kernel on the host, copied it to the usual "/boot/mygreatkernel" and booted the host using it.

3)
Started all my QEMU/KVM VMs with the parameter "-kernel /boot/mygreatkernel", therefore using the same file that my host was using.

Result:
At the beginning everything working perfectly.
After a week or so I started noticing that the VMs were quite slow and after some more time their poor performance was obvious, with a reboot fixing everything until the whole cycle started again a few days/weeks later.
I then started looking a bit deeper into it and I saw that a lot of time was spent on kernel calls.
I then copied the kernel config file (".config") to the VMs themselves, compiled it in the VMs and put them on the host and since I boot using it (every VM has its own kernel-file) everything is fine.

Any clue why?
I could understand that even if the configuration in the ".config" file is the same the kernel compiled in the VMs is different from the one of the host (Xeon CPU) because of the "-native" CFLAG I'm using, but a deterioration of the performance with time should not happen, or?