<div class="gmail_quote">On Wed, Jul 15, 2009 at 3:21 PM, Daniel P. Berrange <<a href="mailto:berrange@redhat.com">berrange@redhat.com</a>> wrote: <blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;"> <div class="im">On Wed, Jul 15, 2009 at 10:12:16AM -0400, Rich Mahn wrote: > I am setting up several VMs using an F11 host. Most of the VMs will > be F11 as well. > > >From earlier discussions on this list I recognize that the best > performance for virtual disks is with the backing storage on the host > being a parititon or LV. Since I want some flexibility I will use > LVs, with virtio disks on most of the VMs. > > Here's my questions: > > 1. Each f11 VM needs three file systems--/boot, root, and swap > (assuming you can call swap a file system). Is it better to create > three LVs (each) on the hosts, and treat it as three separate disks > on the VMs? Or is it better to just create one LV on the host and > then divide up that virtual disk on the VM? </div>There's no point separating /boot & root onto separate virtual disks.</blockquote><div> Really? I have to admit to only being familiar with Xen, so I do not know if this also applies to KVM-style virtualisation. What I have found is that if you use pygrub (a minimal implementation of grub written in Python) and attempt to boot a domain, it will have issues unless /boot is a file. Which has mean that I have made it separate to /, an LVM VG - for the reasons you suggest. Also - within the guest it wants to have a partition table and Xen, at least, will simply have the devices /dev/sda1 and /dev/sda2 but not /dev/sda which means grub (in the guest) complains. Having /boot on a file based disk, with a parition table at the front, means that /dev/sda is provided and that the guest is able to run grub without hacking the device-map file (or temporarily make the device node, since it will show up). So, I'm one of "those people" looking to eventually migrate from Xen to KVM -- is this issue all taken care of by KVM-style virtualisation? Regards, Anand </div></div>