[Melbourne-pm] NYTProf

[Dan Tyrrell]

Hi,

Further to the recent talk (in part) on NYTProf by PauI I noticed that
for certain workloads NYTProf can add significant overheads. See
below. (Note that the process being profiled appeared to be CPU/Memory
bound.)  It is not all bad. I have no idea how this overhead would
compare compare to other profiling options - except that the overhead
was much lower than attempting to find interesting (i.e. slow or
poorly written) parts of the code manually. In this case NYTProf made
it quite clear what part of the code was 'interesting'....it rocks!

[dan at localhost checkpatch]$ time  perl  ./checkpatch.pl --no-tree -q
--file aten2011.c

real	0m8.964s
user	0m8.814s
sys	0m0.025s


[dan at localhost checkpatch]$ time  perl -d:NYTProf ./checkpatch.pl
--no-tree --file aten2011.c

real	2m2.884s
user	2m1.549s
sys	0m0.303s


The discussion that prompted this is here:

http://marc.info/?l=linux-kernel&m=123342823128520&w=2

If anyone is interested I can also post the HTML reports somewhere.

Cheers,

Dan Tyrrell