On Wed, 2013-01-30 at 15:23 -0800, Adam Williamson wrote:
It seems pretty clear from these numbers that slub_debug is the
biggest
offender for me, according to jwb. I didn't put the results in the chart
for some reason, but on the nouveau machine just like the intel one,
rc5.git1 with slub_debug=- as a kernel parameter gives a fast gtkperf
time - in the 2.4-2.5 range - and 'feels' perfectly fine, that's what
I'm running right now.
One thing I should have noted, I think, now I read the mails and see
that other options seem to cause I/O sluggishness for other testers: the
box I'm using has 16GB of RAM and an SSD. So it's quite likely that I
wouldn't perceive I/O sluggishness in regular use unless it got _really_
bad. So my results don't necessarily contradict other people's findings
of I/O sluggishness in relation to options other than slub_debug.
--
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | identi.ca: adamwfedora
http://www.happyassassin.net