Re: RE : Log level and performances

[Buchan Milne <bgmilne@obsidian.co.za>]

On Mon, 19 Apr 2004, BAILLEUX Benoît FTRD/DMI/CAE wrote:

> Thank you for the trick. I'm not very used to play with that kind of tweak, and I didn't know that possibility.
> With logfile written in async mode, the throughput is really better.
> In the same condition, I had 29 op/s with sync'ed syslog's file. I now have 1890 op/s with async file.
> 
> So the results are now :
>  - no log              : 2540 op/s (cool !)
>  - loglevel=256, sync  : 29   op/s (uh !)
>  - loglevel=256, async : 1890 op/s (not bad ...)
>  - slapd -d 256        : 2577 op/s (higher than forseen)
> 
> Note for Quanah :
> For my tests, I use one physical disk for my BDB database, and one *other* for BDB logs and Syslog files.
> However, the file produced by the command line :
>   ./slapd -d 256 <other options> > ../my_log_file
> Is on the disk with the database.
> 
> This is why I don't understand why throughput is slightly better (than 
> slapd without any log) when looging to a standard file. But I suppose 
> that it is more related to the system than to OL ...

Well, are you starting the service up identically (besides the logging 
differences)?

For example, you may be starting slapd in other cases with an init script, 
which may:
-have slapd run as another user
-start slapd with a nice level
-start slapd with slightly different configuration (ie serve on ldaps too)
-start slurpd also

So, I would suggest you either ensure that you start slapd identically in 
the 'slapd -d 256' case (by checking te command line slapd is normally 
started with) or hack the init script to pass the debug level too. If you 
don't, you are not comparing the same thing.

Regards,
Buchan