Re: [Help-gnunet] downloads and CPU

[Christian Grothoff]

On Monday 14 February 2005 02:32, Brent Miller wrote:
> Christian Grothoff wrote:
> >Do you have similar data for gnunet-download?
> >
> >Christian
>
> Attached are the gprof output from gnunet-download and gnunet-gtk with
> the stats window closed and the search aborted.
>
> What's interesting is that with gnunet-download, I can download smaller
> files with out a problem, but as the files get larger gnunet-download
> uses up more and more cpu. If I download a file around a couple of megs,
> everything is okay, if I try and download a 700MB file like the Knoppix
> image I indexed, my computer gets brought to a standstill and the
> download slows to 1-2 kbps, even though it's indexed on my computer. The
> gprof output of gnunet-download is from a 50 meg test file that I have
> indexed locally, which slowed things down substantially, but didn't
> quite bring it to a crawl.
>
> Hope this is useful.

Not really: the traces are clearly only giving numbers for the first thread 
(it is a problem with gprof and threads that I was not aware of, see below).  
In either case, the total runtime accounted for in the traces is miniscule:  
the "run" method taking "100%" of the execution time accounts for a total of 
40ms of CPU consumption -- which is hardly noteworthy.  The same thing 
applies for the gnunet-gtk data: we see certain functions pop up just because 
they are in thread 1, not because they even take a lot of time. 

Anyway, there is a way to get _good_ data: 

http://sam.zoy.org/writings/programming/gprof.html

If you compile the program given there (instructions are also there -- in the 
comment inside of the code) and preload it, you should get useful profiling 
data.

Christian