[Gllug] Virtual disk allocation advice requested

Richard Jones rich at annexia.org
Mon Jun 30 09:39:34 UTC 2008


On Sun, Jun 29, 2008 at 08:57:48PM +0100, David L Neil Mailing list a/c wrote:
> 2 thought one kept /root separate so that even if some other part of the
> file system was jammed up, root could still log-on (same rationale as /var)

Most filesystems reserve extra blocks for the root user anyway, so
they will always be able to log in even if the filesystem is full
(ie. "full" for normal users).

> =I did however neglect to say that as the processor is a pre-Pacifica
> AMD, I'm talking "paravirtualised". I don't know if this makes a
> difference in any of these points (apologies if does).

Shouldn't make any difference.

> =I will give ext3 a try BUT (and this is a double-duh!) in my notes I
> have the following:
> recommendation: PV mode = non-journalling file system, such as ext2.
> If VM crashes/disorderly shtdn, pending trans cannot rec from jnl.
> Set VM boot own pn, non-jnl fs=ext2.
> (sorry, not clear where I read this)

Yes it's true that LVM doesn't properly pass requests to synch the
journal through to the underlying block device.  There's a lot more
about this issue and how it probably won't affect you here:
http://lwn.net/Articles/283161/

> >> 4 I haven't figured out where/when/how one allocates space within a
> >> DomU. Does/should one separate out space allocated to logs, the root
> >> user, etc, as we were advised in 'the good old days' or has such gone by
> >> the board/become irrelevant in the brave new virtual world?
> > I usually use a single root partition for guests (/dev/xvda1 = /boot,
> > /dev/xvda2 = /)
> 
> =is there a particular reason why you don't worry about logs exceeding
> the space available, or some such doomsday scenario - eg you actually
> monitor your system properly?
> (are probably religious about backups too...bless you my son!)

I (now) use virt-df to monitor my guests to look for ones which are
about to run out of disk space.

> =Logic says that if use NFS, eg for /home within any VM and 'house' the
> NFS server within another VM, that at boot time Xen will not guarantee
> that one domain will be up and running prior to another being started
> and thus the situation may produce a time-race condition.

Yup, this is another reason not to use a server in a guest (although
exactly the same problem may happen with an NFS server in the dom0 --
how do you know it is serving requests before the guests start up?).
In any case, NFS is engineered to deal with this situation so you
don't need to worry about it, but take a close look at the 'bg' option
mentioned in the nfs(5) manual page.

> =Contrarily: the advice is to keep everything possible out of Dom0. All
> very well until I get an errmsg telling me that the manager won't run
> without a graphics system (see 'confession' later) so I ended up loading
> Gnome (and it wouldn't let me try to avoid loading the whole kitchen
> sink). Now I think about it, wonder if just X would have been enough???

You can run X applications remotely.

Rich.

-- 
Richard Jones
Red Hat
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug




More information about the GLLUG mailing list