Skip to contentSkip to conversation

pmap revisited

DISCUSSION Samium Gromoff 2 participants 2 messages Sep 3, 2001 – Sep 4, 2001

lore
Share & tools
lore
#0 lore raw

Ashes on my head guys...
Gotta wrong results in my previous perftest... (slightly different
environments), so these are to be sure that on low VM load there isnt any significant difference...

Here are new and revisited. Actually i maked sure that the environments
differs only in kernels...

Bonus: two bugs! :)
1. Quintela`s (shmtest of memtest) and pmap{2,3} == 100% instant deadlock plain ac12 demonstrates ignorance.
2. Swapoff oops 100% - only in pmap3! (okay, swapoff of reiserfs to be strict, but i think that doesnt actually matters) swapoff oops will be in next mail.

Revisited results:

time find / -xdev - done 5 times
pmap 3

real    1m5.175s
real    1m4.699s
real    1m3.579s
============================
pmap 2
real    1m5.039s
real    1m4.779s
real    1m4.506s
============================
plain
real    1m4.820s
real    1m4.433s
real    1m4.285s

* fillmem == (fillmem of memtest)
* dont count on differences - they are quite flaky, one is only known:
there are no much difference anyways...
time fillmem - done 7 times, still giving strange results sometimes... pmap 3

real    1m2.709s
real    1m2.417s
real    1m1.371s
real    1m1.241s
real    1m0.235s
(1m3.~, 1m4.5~) - add results
============================
pmap 2
real    1m5.294s
real    1m5.169s
real    1m4.431s
real    1m3.878s
real    1m3.523s
(1m6.~, 1m0.~, 1m3.355) - add results
============================
plain
real    1m1.570s
real    1m1.063s
real    1m1.007s
real    1m0.201s
real    0m59.677s
#1 lore raw
Quoted 3 lines · Samium Gromoff

Gotta wrong results in my previous perftest... (slightly different
environments), so these are to be sure that on low VM load there isnt any significant difference...

As expected, the current patch only modifies mechanisms and leaves most policy the same. Under some loads there is a difference, but under light VM loads the same-policy-more-info replacement should indeed be pretty similar.

Quoted 3 lines

Bonus: two bugs! :)
1. Quintela`s (shmtest of memtest) and pmap{2,3} == 100% instant deadlock plain ac12 demonstrates ignorance.

I'll try to reproduce that one when I get back home thursday.

Quoted 3 lines

2. Swapoff oops 100% - only in pmap3! (okay, swapoff of reiserfs to be strict, but i think that doesnt actually matters) swapoff oops will be in next mail.

This one is fixed. I'll be sitting inside a big tin can with wings all day tomorrow, but thursday I'll try to post a new version, hopefully with both these bugs fixed.

Thanks for testing the patch and pointing out the bugs!

regards,

Rik

Signature · 5 lines
IA64: a worthy successor to i860.

http://www.surriel.com/		http://distro.conectiva.com/

Send all your spam to [email protected] (spam digging piggy)