Perry, I am glad you brought this up. I 'd be perfectly happy to run the root fs entirely in RAM. In fact that would be ideal for me. It's just I am not quite sure how to use pxeboot to achieve this. I am doing a test now using the tftpboot files created from the iso. As far as I see, the kernel boot options (pxelinux.cfg/default) has: root=live:/ovirt-node-image-2.2.2-1.1.fc16.iso With this I can only pxeboot to the intall screen. What should I use to let the kernel mount the root fs in memory? Something like root=/dev/ram0? David
-----Original Message----- From: Perry Myers [mailto:pmyers@redhat.com] Sent: Wednesday, February 08, 2012 2:54 PM To: Mike Burns Cc: Li, David; users@ovirt.org Subject: Re: [Users] Pxeboot
On 02/08/2012 05:03 PM, Mike Burns wrote:
On Wed, 2012-02-08 at 13:46 -0800, Li, David wrote:
Mike,
If I understand this correctly, today I should be able to pxeboot and nfs mount the root fs from a remote server. Apart from setting up the pxe stuff, I 'd have to populate the ovirt node root fs on the server - perhaps steal it from a disk install. In other words I am concerned about the point from which the kernel starts to execute /init script (in the initramfs) to the point /init is able to mount the final root fs from a remote server.
No, there is no way to set this up currently in ovirt-node. You could install using a remote iscsi lun if you have a hardware iscsi HBA, but there isn't a way to mount a remote nfs share as the root fs.
Supporting a remote NFS share as the root fs isn't even something that requested as an RFE at this point or on the roadmap as far as I'm aware.
It sounds like what you're really looking for is a shared root fs that multiple hosts could use. This is something that we will probably look into eventually, but it's not on the immediate roadmap.
Given that the rootfs of oVirt Node is fairly small and in a truly stateless environment would just run out of system RAM, there's no real reason to try to do a shared NFS based rootfs. It's an unnecessary complication I think, if the end goal is to move to truly stateless.
For larger systems where the rootfs is on the order of GB's, shared root may make more sense.