At 09.18 04/11/01 -0800, Linus Torvalds wrote:
>
>On Sun, 4 Nov 2001, Lorenzo Allegrucci wrote:
>>
>> I begin with the last Linus' kernel, three runs and kswapd CPU
>> time appended.
>
>It's interesting how your numbers decrease with more swap-space. That,
>together with the fact that the "more swap space" case also degrades the
>second time around seems to imply that we leave swap-cache pages around
>after they aren't used.
>
>Does "free" after a run has completed imply that there's still lots of
>swap used? We _should_ have gotten rid of it at "free_swap_and_cache()"
>time, but if we missed it..

lenstra:~/src/qsort> free
             total       used       free     shared    buffers     cached
Mem:        255984      16760     239224          0       1092       8008
-/+ buffers/cache:       7660     248324
Swap:       195512          0     195512
lenstra:~/src/qsort> time ./qsbench -n 90000000 -p 1 -s 140175100
70.590u 7.640s 2:31.06 51.7%    0+0k 0+0io 19036pf+0w
lenstra:~/src/qsort> free
             total       used       free     shared    buffers     cached
Mem:        255984       6008     249976          0        100       1096
-/+ buffers/cache:       4812     251172
Swap:       195512       5080     190432

and with more swap..

lenstra:~/src/qsort> free
             total       used       free     shared    buffers     cached
Mem:        255984      13488     242496          0        532       5360
-/+ buffers/cache:       7596     248388
Swap:       390592          0     390592
lenstra:~/src/qsort> time ./qsbench -n 90000000 -p 1 -s 140175100
70.180u 7.650s 2:43.22 47.6%    0+0k 0+0io 21019pf+0w
lenstra:~/src/qsort> free
             total       used       free     shared    buffers     cached
Mem:        255984       6596     249388          0        108       1116
-/+ buffers/cache:       5372     250612
Swap:       390592       5576     385016
lenstra:~/src/qsort> time ./qsbench -n 90000000 -p 1 -s 140175100
71.030u 7.040s 2:49.45 46.0%    0+0k 0+0io 22734pf+0w
lenstra:~/src/qsort> free
             total       used       free     shared    buffers     cached
Mem:        255984       8808     247176          0        108       1152
-/+ buffers/cache:       7548     248436
Swap:       390592       7948     382644


>What happens if you make the "vm_swap_full()" define in <linux/swap.h> be
>unconditionally defined to "1"?

lenstra:~/src/qsort> free
             total       used       free     shared    buffers     cached
Mem:        256000      16772     239228          0       1104       8008
-/+ buffers/cache:       7660     248340
Swap:       195512          0     195512
lenstra:~/src/qsort> time ./qsbench -n 90000000 -p 1 -s 140175100
70.530u 7.290s 2:33.26 50.7%    0+0k 0+0io 19689pf+0w
lenstra:~/src/qsort> free
             total       used       free     shared    buffers     cached
Mem:        256000       5132     250868          0        116       1144
-/+ buffers/cache:       3872     252128
Swap:       195512       3748     191764

..and now with 400M of swap:

lenstra:~/src/qsort> free
             total       used       free     shared    buffers     cached
Mem:        256000      13096     242904          0        504       4904
-/+ buffers/cache:       7688     248312
Swap:       390592          0     390592
lenstra:~/src/qsort> time ./qsbench -n 90000000 -p 1 -s 140175100
70.830u 7.100s 2:29.52 52.1%    0+0k 0+0io 18488pf+0w
lenstra:~/src/qsort> free
             total       used       free     shared    buffers     cached
Mem:        256000       4980     251020          0        108       1132
-/+ buffers/cache:       3740     252260
Swap:       390592       3840     386752
lenstra:~/src/qsort> time ./qsbench -n 90000000 -p 1 -s 140175100
70.560u 6.840s 2:28.66 52.0%    0+0k 0+0io 18203pf+0w
lenstra:~/src/qsort> free
             total       used       free     shared    buffers     cached
Mem:        256000       5044     250956          0        108       1112
-/+ buffers/cache:       3824     252176
Swap:       390592       3896     386696

Performace improved and numbers stabilized.

>That should make us be more aggressive
>about freeing those swap-cache pages, and it would be interesting to see
>if it also stabilizes your numbers.
>
>		Linus

I attach qsbench.c