Earth Computing

[Scott Moser]

On Thu, 28 Jan 2010, Etienne Goyer wrote:

> Alvin wrote:
> > I know. It's still the most useful filesystem around, and we want to
> > use the best tools for the job. LVM is not as flexible as ZFS.
>
> The main objection I have against ZFS is that it does not have any
> recovery tool (or so I heard; correct me if I am wrong).  As such, I
> would not trust it in production, but you may know something I don't.
>
>
> >>>>   These run Jaunty because of the above bugs and because of a regression
> >>>>   [bug
> >>>>
> >>>> 224138] "No NFS modules in karmic 32-bit"
> >> Again, not trying to make excuse, and not sure I understand the problem
> >> correctly, but that sounds like an overstatement.  It seems like the
> >> -virtual kernel flavor is missing some modules (including those for
> >> NFS*v4*), but you could just as well use the -generic or -server flavor.
> >>  Or am I misunderstanding something?
> >
> > Yes, virtio. The virtual kernels perform faster. In extreme cases, a
> > calculation could take as much as 6 hours. You will run in stability issues
> > when using the normal kernel. (Our old Solaris8 machines do the same in 6
> > days.)
> > Besides, when the virtual machines were build, we had no choice due to this
> > bug: [kvm guests not using virtio for networking lose network connectivity]
> > https://bugs.launchpad.net/bugs/286101
>
> I do not know.  I have the 2.6.31-17-generic kernel running on karmic,
> and it does have virtio compiled as a module.  I run all my KVM vm using
> either the -generic or -server kernel flavour, and I have not had any
> problem so far.  Are you sure you *have* to use the -virtual kernel?

I'm not completely following the above, maybe you're saying you chose
-generic/-server in the hardy time frame (the bug is from hardy).

In karmic and lucid, the -virtual kernel is a literal subset of
 -generic-pae on i386
 -server on amd64
If you compare checksums they're identical.  The kernel team calls it a
"sub-flavour".

This should absolutely not cause any differences in performance.  If you
see differences, its not related to this.  In karmic and lucid for sure,
the only benefit of -virtual is disk footprint in the guest.

> Someone else can clarify, but I was under the impression the -virtual
> flavour was being phased out.  Is that correct?

-virtual is not going away as far as I am aware.  In fact, the most recent
discussion on -devel was that it might someday need to be a separate
entity ("flavour") than a strict subset of -generic-pae/-server ("sub
flavour").

That, though, will only happen if there is good reason for certain drivers
to be compiled into -virtual that might cause problems in
-server/-generic-pae.