9:12 a.m.
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?
2. Do I get better performance/stability by NOT using lvm on the VMs?
In other words, just partition the virtio disk and create
filesystems directly on the partitions? Or do the wonders of LVM
and virtio already build in any advantage that might give?
3. Do I get the best performance/stability by creating an LV for each
disk I need, and then NOT partitioning it on the VM, but using the
whole disk for a file system. In other words, use vda, not vda1,
vda2, etc for the filesystem. This would give the side effect
being able to easily mount the filesystem on the host machine when
the VM wasn't being run.
4. Lastly (for now) are these issues too miniscule in their effects
that I probably shouldn't even be worrying about them?
Thanks for all the help.
Rich
9:21 a.m.
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?
There's no point separating /boot & root onto separate virtual
disks. You could argue there's a possible advatnage in having
swap separate if you expect to use swap in the guest, and you
have separate disk spindles in the host you could dedicate to
the guest' swap disk. To be honest though this is niche benefit
at best.
THe only separation I'd do is for the OS system disks, vs application
data disks (or /home if you intended to store lots of data there).
this makes it easier to upgrade OS - eg provision a new OS with the
latest disto, and then just move your data disk to the new guest ,
and throw away the old guest.
2. Do I get better performance/stability by NOT using lvm on the VMs?
In other words, just partition the virtio disk and create
filesystems directly on the partitions? Or do the wonders of LVM
and virtio already build in any advantage that might give?
LVM makes is easier to resize guest FS. eg add a second disk
to the guest, format it as a LVM PV and add it to your VG.
That's pretty much only wayto add more capacity on a running
guest.
3. Do I get the best performance/stability by creating an LV for each
disk I need, and then NOT partitioning it on the VM, but using the
whole disk for a file system. In other words, use vda, not vda1,
vda2, etc for the filesystem. This would give the side effect
being able to easily mount the filesystem on the host machine when
the VM wasn't being run.
Anaconda will refuse to install onto a raw disk, it mandates
partitioning. Second point is that not using a partition table
can cause unexpected problems - eg vgscan in the host may
accidentally see & activate stuff from your guests' disks
with predictable goodbye-data results if your guest is active.
4. Lastly (for now) are these issues too miniscule in their effects
that I probably shouldn't even be worrying about them?
That's certainly my opinion. Sure you get some performance
but you loose the great administrative flexibility of LVM.
Daniel
--
|: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :|
|: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :|
|: http://autobuild.org -o- http://search.cpan.org/~danberr/ :|
|: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|
10:34 a.m.
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?
There's no point separating /boot & root onto separate virtual
disks. You could argue there's a possible advatnage in having
swap separate if you expect to use swap in the guest, and you
have separate disk spindles in the host you could dedicate to
the guest' swap disk. To be honest though this is niche benefit
at best.
THe only separation I'd do is for the OS system disks, vs application
data disks (or /home if you intended to store lots of data there).
this makes it easier to upgrade OS - eg provision a new OS with the
latest disto, and then just move your data disk to the new guest ,
and throw away the old guest.
2. Do I get better performance/stability by NOT using lvm on the VMs?
In other words, just partition the virtio disk and create
filesystems directly on the partitions? Or do the wonders of LVM
and virtio already build in any advantage that might give?
LVM makes is easier to resize guest FS. eg add a second disk
to the guest, format it as a LVM PV and add it to your VG.
That's pretty much only wayto add more capacity on a running
guest.
3. Do I get the best performance/stability by creating an LV for each
disk I need, and then NOT partitioning it on the VM, but using the
whole disk for a file system. In other words, use vda, not vda1,
vda2, etc for the filesystem. This would give the side effect
being able to easily mount the filesystem on the host machine when
the VM wasn't being run.
Anaconda will refuse to install onto a raw disk, it mandates
partitioning. Second point is that not using a partition table
can cause unexpected problems - eg vgscan in the host may
accidentally see & activate stuff from your guests' disks
with predictable goodbye-data results if your guest is active.
4. Lastly (for now) are these issues too miniscule in their effects
that I probably shouldn't even be worrying about them?
That's certainly my opinion. Sure you get some performance
but you loose the great administrative flexibility of LVM.
Daniel
--
|: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :|
|: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :|
|: http://autobuild.org -o- http://search.cpan.org/~danberr/ :|
|: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|
10:56 a.m.
==<snip>==
> 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?
There's no point separating /boot & root onto separate virtual
disks. You could argue there's a possible advatnage in having
swap separate if you expect to use swap in the guest, and you
have separate disk spindles in the host you could dedicate to
the guest' swap disk. To be honest though this is niche benefit
at best.
THe only separation I'd do is for the OS system disks, vs application
data disks (or /home if you intended to store lots of data there).
this makes it easier to upgrade OS - eg provision a new OS with the
latest disto, and then just move your data disk to the new guest ,
and throw away the old guest.
Yes. I planned on doing a bit of this anyhow. Testing some applications
under different configuration and the like.
> 2. Do I get better performance/stability by NOT using lvm on the
VMs?
> In other words, just partition the virtio disk and create
> filesystems directly on the partitions? Or do the wonders of LVM
> and virtio already build in any advantage that might give?
LVM makes is easier to resize guest FS. eg add a second disk
to the guest, format it as a LVM PV and add it to your VG.
That's pretty much only wayto add more capacity on a running
guest.
I hadn't thought about that. With abundant disk space on the host,
I can also make the second disk large enough to migrate the first
PV on the VM to the second, remove the first pv, and then still
only use one LV from the host.
==<snip>==
> 4. Lastly (for now) are these issues too miniscule in their
effects
> that I probably shouldn't even be worrying about them?
That's certainly my opinion. Sure you get some performance
but you loose the great administrative flexibility of LVM.
Daniel
Sounds good. Thanks for your help.
Rich
11:42 a.m.
----snip snip ----
AARRGGHH. Sorry about the duplicate. Hit the wrong
escape-meta-alt-control-shift keys when trying to
reply.
-- rich
12:23 p.m.
On Wed, Jul 15, 2009 at 3:21 PM, Daniel P. Berrange <berrange(a)redhat.com>wrote:
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?
There's no point separating /boot & root onto separate virtual
disks.
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
5422
days inactive
5422
days old
5 comments
3 participants
participants (3)
-
Anand Kumria -
Daniel P. Berrange -
Rich Mahn