On 03/04/2014 01:23 AM, Joonsoo Kim wrote:
> On Mon, Mar 03, 2014 at 12:02:00PM +0100, Vlastimil Babka wrote:
>> On 02/14/2014 07:53 AM, Joonsoo Kim wrote:
>>> changes for v2
>>> o include more experiment data in cover letter
>>> o deal with vlastimil's comments mostly about commit description on 4/5
>>>
>>> This patchset is related to the compaction.
>>>
>>> patch 1 fixes contrary implementation of the purpose of compaction.
>>> patch 2~4 are for optimization.
>>> patch 5 is just for clean-up.
>>>
>>> I tested this patchset with stress-highalloc benchmark on Mel's mmtest
>>> and cannot find any regression in terms of success rate. And I find
>>> much reduced(9%) elapsed time.
>>>
>>> Below is the average result of 10 runs on my 4GB quad core system.
>>>
>>> compaction-base+ is based on 3.13.0 with Vlastimil's recent fixes.
>>> compaction-fix+ has this patch series on top of compaction-base+.
>>>
>>> Thanks.
>>>
>>>
>>> stress-highalloc
>>> 3.13.0 3.13.0
>>> compaction-base+ compaction-fix+
>>> Success 1 14.10 15.00
>>> Success 2 20.20 20.00
>>> Success 3 68.30 73.40
>>>
>>> 3.13.0 3.13.0
>>> compaction-base+ compaction-fix+
>>> User 3486.02 3437.13
>>> System 757.92 741.15
>>> Elapsed 1638.52 1488.32
>>>
>>> 3.13.0 3.13.0
>>> compaction-base+ compaction-fix+
>>> Minor Faults 172591561 167116621
>>> Major Faults 984 859
>>> Swap Ins 743 653
>>> Swap Outs 3657 3535
>>> Direct pages scanned 129742 127344
>>> Kswapd pages scanned 1852277 1817825
>>> Kswapd pages reclaimed 1838000 1804212
>>> Direct pages reclaimed 129719 127327
>>> Kswapd efficiency 98% 98%
>>> Kswapd velocity 1130.066 1221.296
>>> Direct efficiency 99% 99%
>>> Direct velocity 79.367 85.585
>>> Percentage direct scans 6% 6%
>>> Zone normal velocity 231.829 246.097
>>> Zone dma32 velocity 972.589 1055.158
>>> Zone dma velocity 5.015 5.626
>>> Page writes by reclaim 6287 6534
>>> Page writes file 2630 2999
>>> Page writes anon 3657 3535
>>> Page reclaim immediate 2187 2080
>>> Sector Reads 2917808 2877336
>>> Sector Writes 11477891 11206722
>>> Page rescued immediate 0 0
>>> Slabs scanned 2214118 2168524
>>> Direct inode steals 12181 9788
>>> Kswapd inode steals 144830 132109
>>> Kswapd skipped wait 0 0
>>> THP fault alloc 0 0
>>> THP collapse alloc 0 0
>>> THP splits 0 0
>>> THP fault fallback 0 0
>>> THP collapse fail 0 0
>>> Compaction stalls 738 714
>>> Compaction success 194 207
>>> Compaction failures 543 507
>>> Page migrate success 1806083 1464014
>>> Page migrate failure 0 0
>>> Compaction pages isolated 3873329 3162974
>>> Compaction migrate scanned 74594862 59874420
>>> Compaction free scanned 125888854 110868637
>>> Compaction cost 2469 1998
>>
>> FWIW, I've let a machine run the series with individual patches applied
>> on 3.13 with my compaction patches, so 6 is the end of my series and 7-11 yours:
>> The average is of 10 runs (in case you wonder how that's done, the success rates are
>> calculated with a new R support that's pending Mel's merge; system time and vmstats
>> are currently a hack, but I hope to add R support for them as well, and maybe publish
>> to github or something if there's interest).
>
> Good! I have an interest on it.
Most of the support is now in mmtests' master branch on github. The system time
and vmstats is a hack with hardcoded number of iterations (now 10) that I attach
at the end of the mail.
You need R installed and run compare like this:
compare-kernels.sh --R --iterations 10
>
>>
>> Interestingly, you have a much lower success rate and also much lower compaction cost
>> and, well, even the benchmark times. Wonder what difference in config or hw causes this.
>> You seem to have THP disabled, I enabled, but that would be weird to cause this.
>
> My 10 runs are continuous 10 runs without reboot. It makes compaction success
> rate decline on every trial and therefore average result is so low than yours.
> I heard that you did 10 runs because of large stdev, so I thought that continuous 10 runs
> also can makes the result reliable. Therefore I decided this method although it is not
> proper method to get the average. I had to notify about it. If it confuses you,
> sorry about that.
>
> Anyway, noticeable point of continuous 10 runs is that success rate decrease continuously
> and significantly. I attach the rate of success 3 on every trial on below.
>
> Base
> % Success: 80
> % Success: 60
> % Success: 76
> % Success: 74
> % Success: 70
> % Success: 68
> % Success: 66
> % Success: 65
> % Success: 63
> % Success: 61
>
>
> Applied with my patches
> % Success: 81
> % Success: 78
> % Success: 75
> % Success: 74
> % Success: 71
> % Success: 72
> % Success: 73
> % Success: 70
> % Success: 70
> % Success: 70
>
> It means that memory is fragmented continously. I didn't dig into this problem, but
> it would be good subject to investigate.
You're right, I see that too. And the reduction of work done by compaction,
especially within first 3 iterations, is interestingly large.
The number of reclaimable and unmovable pageblocks indeed slightly grows,
although not as much as to translate directly into the success rates.
stress-highalloc
test test test test test test test test test test
1 2 3 4 5 6 7 8 9 10
Success 1 49.00 ( 0.00%) 51.00 ( -4.08%) 51.00 ( -4.08%) 62.00 (-26.53%) 53.00 ( -8.16%) 57.00 (-16.33%) 55.00 (-12.24%) 58.00 (-18.37%) 58.00 (-18.37%) 46.00 ( 6.12%)
Success 2 55.00 ( 0.00%) 64.00 (-16.36%) 62.00 (-12.73%) 62.00 (-12.73%) 56.00 ( -1.82%) 57.00 ( -3.64%) 58.00 ( -5.45%) 59.00 ( -7.27%) 57.00 ( -3.64%) 55.00 ( 0.00%)
Success 3 85.00 ( 0.00%) 81.00 ( 4.71%) 77.00 ( 9.41%) 77.00 ( 9.41%) 75.00 ( 11.76%) 74.00 ( 12.94%) 74.00 ( 12.94%) 73.00 ( 14.12%) 73.00 ( 14.12%) 71.00 ( 16.47%)
test test test test test test test test test test
1 2 3 4 5 6 7 8 9 10
User 5696.69 5713.20 5668.87 5412.19 5425.39 5357.67 5402.08 5345.72 5575.73 5473.66
System 1008.77 1028.72 1025.66 1015.87 1023.82 1023.45 1023.26 1019.52 1026.88 1028.05
Elapsed 2253.63 2285.87 2531.58 2198.46 2220.21 2199.75 2237.43 2266.97 2724.87 2288.90
test test test test test test test test test test
1 2 3 4 5 6 7 8 9 10
Minor Faults 248584347 241444309 238931712 229339487 234134233 234113609 236172742 235955684 235274197 234072299
Major Faults 768 6273 5624 6146 7638 7038 7279 8475 6837 8948
Swap Ins 24 17320 12557 11956 31729 28434 29228 43494 29528 41212
Swap Outs 958 94106 106680 81672 147012 95169 133733 142124 133492 157440
Direct pages scanned 213760 42549 118178 22266 4818 3634 51636 23166 17319 40327
Kswapd pages scanned 2137836 3449877 3500913 3909294 3343227 3800324 3840356 3035274 3048595 3192582
Kswapd pages reclaimed 2134695 2231079 2536612 2238583 2232883 2219601 2180405 2324381 2331881 2276682
Direct pages reclaimed 213422 42304 117833 22134 4711 3562 51551 23088 17240 40281
Kswapd efficiency 99% 64% 72% 57% 66% 58% 56% 76% 76% 71%
Kswapd velocity 948.619 1509.218 1382.896 1778.197 1505.816 1727.616 1716.414 1338.912 1118.804 1394.811
Direct efficiency 99% 99% 99% 99% 97% 98% 99% 99% 99% 99%
Direct velocity 94.851 18.614 46.682 10.128 2.170 1.652 23.078 10.219 6.356 17.619
Percentage direct scans 9% 1% 3% 0% 0% 0% 1% 0% 0% 1%
Zone normal velocity 329.358 931.485 794.284 1165.576 922.453 1122.753 1146.195 711.711 594.119 801.484
Zone dma32 velocity 714.112 596.348 635.294 622.749 585.533 606.515 593.297 637.420 531.040 610.945
Zone dma velocity 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Page writes by reclaim 958.000 142306.000 147653.000 154673.000 168131.000 120184.000 157196.000 158862.000 155427.000 178478.000
Page writes file 0 48200 40973 73001 21119 25015 23463 16738 21935 21038
Page writes anon 958 94106 106680 81672 147012 95169 133733 142124 133492 157440
Page reclaim immediate 227 978397 717495 1393234 747156 1277758 1307521 315722 396171 528904
Sector Reads 3146168 3275924 4886280 3301644 3328564 3341776 3357484 3431016 3375072 3329768
Sector Writes 12552716 12767728 12971248 12339364 12829408 12581020 12810572 12822476 12836888 12906372
Page rescued immediate 0 0 0 0 0 0 0 0 0 0
Slabs scanned 1857664 2226176 2273280 2238464 2250752 2269184 2297856 2229248 2216960 2219008
Direct inode steals 10976 10215 17701 2688 19256 8041 3827 11226 10767 13599
Kswapd inode steals 58552 341409 352875 362913 406943 392956 454265 326993 310959 341106
Kswapd skipped wait 0 0 0 0 0 0 0 0 0 0
THP fault alloc 97 59 96 8 58 54 58 68 68 54
THP collapse alloc 655 416 616 396 396 393 402 386 396 387
THP splits 6 5 18 12 10 11 9 12 11 7
THP fault fallback 0 0 0 0 0 0 0 0 1 0
THP collapse fail 11 21 13 21 21 20 22 18 20 21
Compaction stalls 5851 6125 6925 5572 6183 6064 6224 6069 6196 6861
Compaction success 1163 1811 2144 1783 1747 1886 1833 1823 1781 1748
Compaction failures 3250 2200 2195 1800 1754 1421 1834 1521 1929 1753
Page migrate success 6110592 3508332 3093592 2740374 2091309 1400876 2208633 1516013 2223733 1699633
Page migrate failure 0 0 0 0 0 0 0 0 0 0
Compaction pages isolated 15360926 10185640 9900328 8228257 6719820 5195540 6677838 4841171 6708180 5479883
Compaction migrate scanned 349981387 300209536 297909273 286215718 268660759 256103243 285198850 241549363 287544164 261619750
Compaction free scanned 556386189 260105925 296420166 196203596 148614315 98672618 174033818 82384264 165998395 123011890
Compaction cost 9084 5936 5484 5004 4179 3345 4415 3356 4448 3699
NUMA alloc hit 167473642 162715095 160968619 154496168 157513861 157598611 159117516 158854501 158279702 157440309
NUMA alloc miss 0 0 0 0 0 0 0 0 0 0
NUMA interleave hit 0 0 0 0 0 0 0 0 0 0
NUMA alloc local 167473642 162715095 160968619 154496168 157513861 157598611 159117516 158854501 158279702 157440309
NUMA page range updates 0 0 0 0 0 0 0 0 0 0
NUMA huge PMD updates 0 0 0 0 0 0 0 0 0 0
NUMA PTE updates 0 0 0 0 0 0 0 0 0 0
NUMA hint faults 0 0 0 0 0 0 0 0 0 0
NUMA hint local faults 0 0 0 0 0 0 0 0 0 0
NUMA hint local percent 100 100 100 100 100 100 100 100 100 100
NUMA pages migrated 0 0 0 0 0 0 0 0 0 0
AutoNUMA cost 0 0 0 0 0 0 0 0 0 0
test test test test test test test test test test
1 2 3 4 5 6 7 8 9 10
Mean sda-avgqz 62.27 64.24 67.82 68.47 64.52 74.36 68.65 66.98 52.45 61.41
Mean sda-await 356.96 384.46 396.37 426.02 384.43 447.38 409.69 384.68 315.29 357.33
Mean sda-r_await 43.62 44.58 40.66 49.54 45.40 47.44 48.68 41.98 35.12 39.42
Mean sda-w_await 705.15 813.62 1136.46 1029.17 834.44 1127.80 971.66 914.88 712.12 784.44
Max sda-avgqz 291.35 286.96 162.48 265.33 161.77 240.23 196.98 203.30 162.28 283.29
Max sda-await 1258.88 1680.93 2309.73 1566.46 1711.69 1948.61 1381.23 1986.32 1583.03 1737.79
Max sda-r_await 274.27 185.52 125.71 272.05 340.00 332.00 312.00 188.16 159.23 173.33
Max sda-w_await 8174.71 12488.89 13484.06 12931.84 9821.61 10462.28 9182.25 9573.30 12516.77 8842.26
> Thanks.
>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@xxxxxxxxx. For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
>
------8<------
commit 596bad5b6505b8b606c197a8ee2af8a74a7edbc6
Author: Vlastimil Babka <vbabka@xxxxxxx>
Date: Mon Dec 16 13:38:36 2013 +0100
HACK: iteration support for MonitorDuration and MonitorMmtestsvmstat
diff --git a/bin/lib/MMTests/MonitorDuration.pm b/bin/lib/MMTests/MonitorDuration.pm
index 1069e83..313eb0a 100644
--- a/bin/lib/MMTests/MonitorDuration.pm
+++ b/bin/lib/MMTests/MonitorDuration.pm
@@ -19,22 +19,33 @@ sub new() {
sub extractReport($$$) {
my ($self, $reportDir, $testName, $testBenchmark) = @_;
- my $file = "$reportDir/tests-timestamp-$testName";
+ my $file;
+ my $i;
+ my ($user, $system, $elapsed);
+ my $iterations = 10;
+
+ for ($i = 1; $i <= $iterations; $i++) {
+ $file = "$reportDir/$i/tests-timestamp-$testName";
open(INPUT, $file) || die("Failed to open $file\n");
while (<INPUT>) {
if ($_ =~ /^time \:\: $testBenchmark (.*)/) {
my $dummy;
- my ($user, $system, $elapsed);
+ my ($useri, $systemi, $elapsedi);
- ($user, $dummy,
- $system, $dummy,
- $elapsed, $dummy) = split(/\s/, $1);
+ ($useri, $dummy,
+ $systemi, $dummy,
+ $elapsedi, $dummy) = split(/\s/, $1);
- push @{$self->{_ResultData}}, [ "", $user, $system, $elapsed];
+ $user += $useri;
+ $system += $systemi;
+ $elapsed += $elapsedi;
}
}
close INPUT;
+ }
+
+ push @{$self->{_ResultData}}, [ "", $user / $iterations, $system / $iterations, $elapsed / $iterations];
}
1;
diff --git a/bin/lib/MMTests/MonitorMmtestsvmstat.pm b/bin/lib/MMTests/MonitorMmtestsvmstat.pm
index 7bd3a21..bec917b 100644
--- a/bin/lib/MMTests/MonitorMmtestsvmstat.pm
+++ b/bin/lib/MMTests/MonitorMmtestsvmstat.pm
@@ -167,8 +167,13 @@ sub extractReport($$$$) {
my $elapsed_time;
my %zones_seen;
- my $file = "$reportDir/tests-timestamp-$testName";
+ my $i;
+ my $iterations = 10;
+ my $file;
+
+ for ($i = 1; $i <= $iterations; $i++) {
+ $file = "$reportDir/$i/tests-timestamp-$testName";
open(INPUT, $file) || die("Failed to open $file\n");
while (<INPUT>) {
if ($_ =~ /^test begin \:\: $testBenchmark/) {
@@ -209,9 +214,9 @@ sub extractReport($$$$) {
my ($key, $value) = split(/\s/, $_);
if ($reading_before) {
- $vmstat_before{$key} = $value;
+ $vmstat_before{$key} += $value;
} elsif ($reading_after) {
- $vmstat_after{$key} = $value;
+ $vmstat_after{$key} += $value;
}
if ($key eq "pgmigrate_success") {
$new_compaction_stats = 1;
@@ -222,6 +227,12 @@ sub extractReport($$$$) {
}
}
close INPUT;
+ }
+
+ foreach my $key (sort keys %vmstat_before) {
+ $vmstat_before{$key} /= $iterations;
+ $vmstat_after{$key} /= $iterations;
+ }
# kswapd steal
foreach my $key ("kswapd_steal", "pgsteal_kswapd_dma", "pgsteal_kswapd_dma32",
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>