He1p Required - Please??? (fwd)

Ian Freislich iang at uunet.co.za
Fri Aug 13 10:20:44 UTC 1999


Lloyd Rennie wrote:
> On Fri, 13 Aug 1999, Fabien Tassin wrote:
> > According to Lloyd Rennie:
> > > We're using CNFS, on a ccd array sd2-sd7.  I realise this is
> > > probably not a useful 'report', I'm frankly doubtful it's even
> > > a bug.  But I've run out of things to try, so I'm asking you.
> > > Please let me know what further information, config. details etc.
> > > would be appropriate.
> >
> > It looks like a 'df' bug. I don't know if 'inndf' can handle this.
> > Can we have more information about the OS used and about this ccd
> > array ?

I don't think that there is a bug with df or CNFS that caused this.
I think, unless I've missed something, that Lloyd might have missed
the point of concatenated disk driver.

We run our news spool on a ccd (12*4.2G disks) with no problem:
[proxy05] /var/spool/news # ccdconfig -g
ccd0            128     1       /dev/da18s1e /dev/da19s1e /dev/da20s1e
/dev/da21s1e /dev/da24s1e /dev/da25s1e /dev/da26s1e /dev/da27s1e
/dev/da28s1e /dev/da29s1e /dev/da30s1e /dev/da31s1e

[proxy05] /var/spool/news # df -k
Filesystem  1K-blocks     Used    Avail Capacity  Mounted on
/dev/da0s1a     31743    22173     7031    76%    /
/dev/da0s1f   2743075   765465  1758164    30%    /usr
/dev/da0s1e    508143    80825   386667    17%    /var
/dev/ccd0c   48761816 39163109  5697762    87%    /var/spool/news
procfs              4        4        0   100%    /proc

> The OS is FreeBSD 3.2-19990518-STABLE, running
> inn-STABLE_2_2-1999-06-02_03-02 The ccd.conf file looks like this...
>
> ccd0 65536 none /dev/da1e /dev/da4e /dev/da2e /dev/da5e /dev/da3e
> /dev/da6e /dev/da7e

That's quite a large interleave factor. We set ours to 64k (the
maximum block transfer size), or 128 blocks with good performance
results. This gives the disks time to 'recover' between reads and
writes. Your interleave factor is 32Mbytes which means 32Mb is
written before moving on to the next disk in your stripe.

However, from your mount output, it looks as though you're not
using your stripe at all. Once you've configured your ccd, you need
to newfs and mount it. Don't mount the individual disks in the stripe,
once you've newfs'd the ccd, the individual disks will each have
a part of the file system on the ccd.

Use 'ccdconfig -g' to dump the details of currently configured
ccds.  Once this looks O.K., newfs the ccd (we use the 'c' partition
because we're too lazy to relable the ccd).

I'm afraid you're probably going to have to start again :(

You can correspond with me in private if you need more detailed
help on making your ccd work.

-- 
Ian Freislich
UUNET S.A. (Pty) Ltd


More information about the inn-workers mailing list