[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
it gets slower
After doing a huge amount of disk I/O the other day I noted
that a subsequent full kernel build took *ages*. I haven't
been able to reproduce the full extent of this, but..
Doing a build after boot takes 320 seconds. After
copying two kernel trees and diffing them, an identical
full build takes 336 seconds. Profiling each build says:
Before:
c01120b0 do_page_fault 1453 1.2440
c0271a00 __generic_copy_to_user 1741 29.0167
c0122c6c do_wp_page 1895 2.9609
c012da88 rmqueue 2245 3.7922
c0271a3c __generic_copy_from_user 3593 59.8833
c01260f0 file_read_actor 4490 41.5741
c0123388 do_anonymous_page 11121 89.6855
c01071c0 default_idle 21741 418.0962
00000000 total 71684 0.0469
After:
c0271a00 __generic_copy_to_user 1696 28.2667
c0122c6c do_wp_page 1865 2.9141
c012da88 rmqueue 2226 3.7601
c0271a3c __generic_copy_from_user 3584 59.7333
c01260f0 file_read_actor 4529 41.9352
c0123388 do_anonymous_page 11324 91.3226
c01071c0 default_idle 34807 669.3654
00000000 total 90069 0.0590
This is a 1kHz profile, so we've spent an extra 13 seconds in
default_idle. Doing I/O, presumably. 256 meg dual Celeron.
Why?
Also, we spent 12 seconds in do_anonymous_page and do_wp_page
madly memsetting. Has anyone tried prezeroing some pages in
default_idle(), so they're ready to go?
Yes, I know it'll probably be cachely disadvantageous,
but has anyone actually tried and measured it?
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/