On Monday 24 February 2003 21:36, Martin Konold wrote:
> Am Montag, 24. Februar 2003 02:37 schrieb Josef Weidendorfer:
>
> Hi Josef,
>
> > > The dramatic measurment results from Christian indicate that there is
> > > much more time spent at least on his system.
> >
> > With KDE3.1 (Suse8.0 packages) top shows 1.9 % constant CPU utilization
>
> Top is totally bogus. Please dont use top for such measurements.

So what are you suggesting for meassuring?

> > for kded. strace confirms that this is indeed a row of stats every 0,5
> > seconds. But my estimation of 40 dirs was wrong: There are at least >150
> > dirs watched (Standard KDE desktop dirs + a LOT of additional Suse dirs)
> >
> > :-(
>
> Outch.
>
> Is it really necessary to watch this many dirs with a 0.5 s intervall. I
> have the strong impression that a much longer intervall would also be
> acceptable in 99% of the cases.

The poll interval is adjustable over a global config option. 0,5 s is the 
default poll interval for local files and all KDirWatch uses. Should we 
instead default to a calculated value depending on processor speed?
There's no way for an application using KDirWatch to specify a poll interval. 
Note that this makes only limited sense at all: The system should decide 
itself, whether to use DNOTIFY or FAM or STAT, and when using STAT, it should 
adjust the interval to whether the file is on a NFS share or not.

Perhaps we could introduce a qualitative option: "fast" or "slow" indication 
of changes. This could go very well with a delaying/merging feature of 
DNOTIFY/FAM notifications.

Josef

>
> (Optimize for the common case)
>
> Regards,
> -- martin
>
> Dipl.-Phys. Martin Konold
>
> e r f r a k o n
> Erlewein, Frank, Konold & Partner - Beratende Ingenieure und Physiker
> Nobelstrasse 15, 705