Re: Abysmal performance of media-export on a medium sized collection

[Jens Georg <mail jensge org>]

On Do, 2012-07-19 at 12:31 +0200, Roderich Schupp wrote:

> There's definitely something fishy here as the scan seems to get
> slower and slower towards the end. If I measure the time to process
> one track (from the first open of a certain .mp3 file to the first open
> of the next), it goes from 0.02 sec at 1 minute into the scan
> over 4.86 sec at the 50 minute mark to 7.76 sec towards the end.

Yes, I know that behaviour, but I haven't had time to look deeper into
this. 

> 
> I've attached three snippets of "strace -f -v -ttt rygel" for the processing of
> a single track at the above times. Note that the behaviour at 50
> and 98 minutes is different from that at 1 minute: we see that
> a thread is created (with clone()) that doesn't seem to do anything
> except to wait for a futex.

Rygel doesn't spawn threads on its own so no clue where that comes
from :)

> 
> Note that the whole scan (i.e. walk the filesystem tree,
> extract ID3 tags and stuff the information into a sqlite3 database)
> should only take a couple of minutes, as demonstrated by other
> UPNP servers, e.g. gmediaserver.

Yes, I know and agree totally.