Steve Lord schrieb:
>
> On Fri, 2002-04-12 at 09:04, Simon Matter wrote:
> > Seth Mos schrieb:
> > >
> > > At 14:25 12-4-2002 +0200, =?ISO-8859-2?Q?Libor_Van=ECk?= wrote:
> > > >Hi,
> > > >I'd like to ask 2 small yes/no questions about XFS:
> > > >- does XFS support fs resising?
> > >
> > > Only larger, shrinking is not supported.
> > >
> > > >- is there any way how to have physicaly more (Linux) machines which
> > > >would
> > > >act like one big XFS (probably all driven by one "master" machine which
> > > >would handle I/O requests)?
> > >
> > > There are some other filesystem block layers that can do this but I don't
> > > know if any of them actually work with XFS. I see some trivial test
> > > reports
> > > but I can't remember if any of them was succesful or not.
> >
> > You could use network block devices on several physikal machines and
> > build one big RAID0/1/5 volume with it on the master server. Then put
> > XFS or LVM/XFS on top of it. I tried this once and it worked quite well
> > and fast. If been told you can get better performance than with NFS.
> >
> > -Simon
> >
>
> If you mount one xfs filesystem from several hosts like this then you
> are heading for data corruption very quickly. There is no way to manage
> cache coherency between the machines in this setup, and all the machines
> can end up modifying metadata independently.
This is what I meant:
+----------+ +----------+ +----------+
| Server0 | | Server1 | | Server2 |
| | | | | |
| | | | | |
| /nbd0 | | /nbd1 | | /nbd2 |
| ^ | | ^ | | ^ |
+-----|----+ +-----|----+ +-----|----+
| | |
| | |
| | |
| \-------\ |
| | |
| | |
| | |
| +-------------------|--+ |
| | Big Iron | | |
| | | | |
| | | | |
| | | | |
\----------->/dev/nbd0 | | |
| /dev/nbd1 <-/ | |
| /dev/nbd2 <-----------/
| |
| /dev/md0:RAID5 { |
| /dev/nbd0 |
| /dev/nbd1 |
| /dev/nbd2 } |
| |
| |
| XFS on /dev/md0 |
| |
+----------------------+
Is _this_ dangerous or did you get me wrong? IIRC it has worked perfect
and I didn't see any corruption.
-Simon
>
> The only way to do this now is NFS. CXFS will allow this configuration
> when it is available for Linux. However CXFS is still a ways off, and
> will not be open sourced.
>
> Steve
>
> --
>
> Steve Lord voice: +1-651-683-3511
> Principal Engineer, Filesystem Software email: lord@xxxxxxx
|