On Wed, Mar 06, 2002 at 05:54:51PM -0600, Steve Lord wrote:
> On Wed, 2002-03-06 at 17:48, Matthijs van der Klip wrote:
> > Hi,
> >
> > Could someone tell me what the status of the 'null' issue is?:
> >
> > http://oss.sgi.com/projects/xfs/faq.html#nulls
> >
> >
> > Has any solution or workaround been developed yet? Can we expect a
> > solution in the next official release?
> >
>
> The 2.4.18 kernel in cvs and the patches has some code which removes
> the synchronous transactions from xfs. For the most part these tended
> to be the culprit in that an inode size update was forced out into
> the log long before the data was.
>
> So, while it is still possible, it is less likely than it was.
I guess one can make it even less likely by changing the bdflush buffer timeout
to 5 seconds, at the cost of some performance.
% cat /proc/sys/vm/bdflush
30 500 0 0 500 3000 60 20 0
% echo 30 500 0 0 500 500 60 20 0 >
/proc/sys/vm/bdflush
[3000 -> 500]
-Andi
|