[plug] Don't do that!

[Bernd Felsche]

On Thursday 16 September 2004 12:10, Brad Campbell wrote:
> Bernd Felsche wrote:
> > building a filesystem on them. I'm not mirroring swap, instead
> > assigning equal-priority swap partitions on both drives and letting
> > the kernel handle the striping.

> Why not? You go to the trouble of mirroring all the data, but if
> you lose a disk with some allocated swap space on it the machine
> *will* die a horrid death.

And reboot 120 seconds later, being only able to access half the
swap space and lose the LVm scratch space...  Not much of an issue.
It's not like it's running anything important. I wouldn't provide a
machine to PLUG for anything *really* important - because I can't
afford to do that.  It's simply too expensive.

What is important is that critical data be mirrored because routine
backups are unlikely.

> > I ended up with four filesystems on RAID1, /, /var, /home and /srv.
> > 6GB, 10GB, 10GB and 30GB respectively.  The rest of the space from
> > the two '80GB' discs has been allocated to /scratch, unmirrored and
> > under LVM.

> > Once I started installation, booting off a USB stick and using media
> > on the server, I noted the error of my ways.  Each md driver
> > commenced mirror recovery to the second drive.  Fortunately, the
> > drivers noticed that they were about to collide on the bus and
> > kindly waited for another to complete.  There was no deadly embrace
> > so it all sort of worked; very slowly; over 10 hours!

> If you mark one disk as removed then add the partitions manually that
> won't happen and it will sequence the rebuilds.

Unfortunately, the installation process gave me no chance to do
that. The rebuilds were sequenced anyway; but remember that it's
56GB to be copied... so it'll take considerable time anyway.

> > Load averages were in excess of 6 with 98% IO wait.
> >
> > In part, that's due to the single IDE channel with master and slave
> > mirroring.

> Oooh. What happens if one drive dies and takes out the IDE
> channel? (Not an unheard of occurence with IDE disks).

New mainboard.

I've seen about 2 dozen IDE hard drive failures in the past decade.
No controllers taken out.

> In addition, by using a single IDE channel you significantly
> knobble the IO on the machine as all writes are duplicated and
> have to contend for the one channel.

After counting the IDE channels (twice), I got to a grand total of
one on each count. There's another socket there for a floppy but
mirroring to a floppy seems pretty pointless in the real world.

-- 
/"\ Bernd Felsche - Innovative Reckoning, Perth, Western Australia
\ /  ASCII ribbon campaign | I'm a .signature virus!
 X   against HTML mail     | Copy me into your ~/.signature
/ \  and postings          | to help me spread!