[rhelv6-beta-list] My first experiences with RHEL6 beta

Nico Kadel-Garcia nkadel at gmail.com
Sat Jun 12 17:40:48 UTC 2010


This was a long comment from Jon Masters. I'll focus on one point.

On Thu, Jun 10, 2010 at 2:03 PM, Jon Masters <jcm at redhat.com> wrote:
> On Thu, 2010-06-10 at 17:19 +0800, John Summerfield wrote:
>
>> First, I did a text install. I was very surprised how many pointless
>> questions it asked, and how many sensible ones it didn't ask. Like, "Do
>> you wan swap?"
>
> Can you let us know if there was a reason that you could not do a vnc
> install? This will give you a lot of flexibility during installation.
>
>> I do not want swap. I especially do not want 2 Gbytes of swap on a test
>> virtual machine.
>
> It can be a little frustrating to use virtual disk space for swap
> volumes, but it is highly recommended in general by the Linux community
> that you use a swap volume. As has been pointed out, LVM really does
> mitigate a lot of the file-vs-volume issues. The kernel really likes
> having swap because it is able to better manage cacheing and virtual
> memory with the option to page stuff out to disk. Even under virt.,
> there are many good reasons to have swap around. There is separate work
> ongoing upstream to reduce double-cacheing type situations, but they
> don't pertain to using swap space with a virtual machine.

Nonsense. While LVM is useful for re-allocating disk space for
alternative partition layouts, and its snapshot capability is very
useful in some environments, and it can handle this reasonably
gracefully, it does absolutely nothing for resizing the filesystems.
That requires more sophisticated tools, which so far as I can tell
remain poorly integrated with Linux. There are a few tools that take a
shot at thus, such as the Gnome parted utility (gparted, often
mis-identified as GNU parted). That's a very powerful and flexible
tool, and I'd have loved to use it for the "4096 byte block virtual
image alignment" problem, but it doesn't have resolution down to
sectors, only to cylinders, so I couldn't use it.

The LVM tools also suffer from rather inconsistent formats and
syntactical usage. Mounted disk partitions show up as
/dev/[volgroup]/[logvol]. But that "file" is actually a symlink to
/dev/mapper/volgroup-logvol. The result is that, for example, such
partitions used for swap don't show up as "/dev/[volgrouup]/[logvol]",
they show up with the actual /dev/mapper/* name, and the contents of
/proc/swap have to be parsed appropriately to go with that.

LVM is also a bad, bad, bad idea for virtualized environments, which
can use the virtualization host's tools to generate snapshots and for
which probing the filesystem of the guest for analysis or data
recovery is seriously encumbered by needing to manipulate LVM inside
the guest's image files. And ye gods, if two guests have identical
naming schemes you're in for a world of hurt.


>> When I do want swap, I usually want a swap file
>
> I am not aware of many distributions that support this mode of
> operation, and I can't remember the last time I was asked about it (vs.
> some unusual nbd-type-swap) but I do think there's some merit in having
> file based swap in the general longer term if we get good low-memory
> notification and reservation support upstream so we can have the system
> dynamically adjust the swap size. But again, I am not aware of anyone
> really asking for that, I'm just interested for academic reasons.

For virtualized guest images, it's common to use a distinct swap disk
image so that the swap can go on separate server disk, space that is
not backed up or used for virtual guest snapshots. The last round of
tests I did with virt-manager under RHEL 5, unfortunately, didn't
permit this at installation time from the GUI. I had to discard
virt-manager to do this from the command line, which I considered
unreasonable. (GUI's should offer the same options as the command
lines, and vice versa.)




More information about the rhelv6-beta-list mailing list