[Gllug] Slow machine

[Daniel P. Berrange]

On Sun, Sep 04, 2005 at 09:52:29AM +0100, Russell Howe wrote:
> Still, I don't see that even a data journalling filesystem should
> affect read performance too much...

Unless you mount with 'noatime', then potentially every single read()
triggers a write in the journal to update atime metadata. From 'man 2 
stat':

  "The  field  st_atime  is  changed  by file accesses, e.g. by execve(2),
   mknod(2), pipe(2), utime(2) and read(2)  (of  more  than  zero  bytes)."

Of course these aren't large write requests, but depending on workload
they could disturb things enough to be a noticable performance hit.

> If you're doing things like video capture, where you're writing data
> sequentially and needing to happen at a guaranteed rate, then
> having fancy things like journalling getting in the way is probably a
> bad idea, but this is what O_DIRECT's for, right? :)
> 
> Modern hard drives do all sorts of fancy stuff under the hood anyway, so
> I shouldn't think it's possible to guarantee anything about the
> timeliness of data writes, unless you were to use specialist kit 

Nor guarentee anything about the integrit of the data written ;-)

Dan.
-- 
|=-            GPG key: http://www.berrange.com/~dan/gpgkey.txt       -=|
|=-       Perl modules: http://search.cpan.org/~danberr/              -=|
|=-           Projects: http://freshmeat.net/~danielpb/               -=|
|=-   berrange at redhat.com  -  Daniel Berrange  -  dan at berrange.com    -=|
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 240 bytes
Desc: not available
URL: <http://mailman.lug.org.uk/pipermail/gllug/attachments/20050904/8a20e1a4/attachment.pgp>
-------------- next part --------------
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug