>> Test:
>>
>> Run loop of 10 iterations of Linux Test Project's "mtest01 -p80 -w"
>> This test attempts to allocate 80% of virtual memory and write to
>> each page. Simulataneously listen to mp3blaster.
>>
> You should try out 2.4.13pre2.aa1..
Elapsed (wall clock) time in seconds is much better in 2.4.13-pre2aa1:
2.4.12aa1
Averages for 10 mtest01 runs
bytes allocated: 1253362892.8
User time (seconds): 2.099
System time (seconds): 2.823
Elapsed (wall clock) time: 64.109
Percent of CPU this job got: 7.5%
Major (requiring I/O) page faults: 135.2
Minor (reclaiming a frame) faults: 306779.8
2.4.13-pre2aa1
Averages for 10 mtest01 runs
bytes allocated: 1245184000
User time (seconds): 2.050
System time (seconds): 2.874
Elapsed (wall clock) time: 49.513
Percent of CPU this job got: 9.7%
Major (requiring I/O) page faults: 115.6
Minor (reclaiming a frame) faults: 304781.9
Note on mp3blaster audio performance:
2.4.12aa1 got smooth at a lower swpd number than 2.4.13-pre2aa1.
2.4.12aa1 became smooth when swpd hit around 130000-140000 (and above).
2.4.13-pre2aa1 became smooth when swpd hit around 280000-300000 (and above).
2.4.13-pre2aa1 vmstat 1 output. This is an example from one of the 10 iterations.
At the point the vmstat output starts, mp3blaster has been playing smoothly for
about 20 seconds.
r b w swpd free buff cache si so bi bo in cs us sy id
2 2 1 554920 1564 1188 2312 44 10092 432 10188 498 719 3 7 91
0 3 1 565652 2288 1192 2400 16 10248 388 10248 423 604 4 9 87
2 1 1 574848 1628 1196 2388 48 9780 488 9780 449 650 2 5 93
2 1 1 586608 1780 1188 2320 56 10704 400 10704 498 682 2 6 92
2 1 1 603504 1784 1184 2316 0 17536 300 17536 457 626 2 10 88
1 2 1 621424 2552 1180 2316 0 18152 64 18192 463 632 3 14 83
0 2 1 640880 2556 1184 2316 0 18340 72 18340 484 687 2 9 89
1 2 1 660848 2036 1180 2316 0 20156 60 20156 487 657 6 8 86
3 0 1 678768 1788 1184 2320 0 17356 76 17356 488 696 3 13 84
0 2 1 696176 2540 1176 2324 0 19416 60 19416 472 667 4 8 89
2 2 1 716144 1656 1172 2324 0 18060 68 18084 481 678 2 13 85
0 2 1 738160 2172 1176 2316 12 23544 88 23544 566 772 6 7 88
3 0 0 14700 316844 1236 2640 284 2048 656 2048 415 728 24 37 39
Around here, mp3blaster begins to stutter.
2 0 0 14696 51576 1236 2768 64 0 192 0 278 579 49 51 0
1 3 1 26728 3384 1188 2024 612 16232 732 16272 675 710 7 13 80
0 3 1 39312 2512 1196 2076 248 13516 428 13520 414 487 4 2 94
3 1 1 50420 1912 1176 2016 216 11800 356 11800 389 397 4 2 95
1 2 0 66564 2568 1184 2008 288 13844 432 13844 458 484 4 4 92
1 2 1 80204 2036 1180 2000 196 16296 256 16320 509 494 2 7 91
1 1 1 97400 1784 1184 2000 236 16848 308 16848 509 464 2 7 92
1 2 1 116720 1904 1180 2000 120 19076 176 19076 508 429 6 5 88
procs memory swap io system cpu
r b w swpd free buff cache si so bi bo in cs us sy id
0 2 1 134488 3060 1188 2000 152 17124 212 17124 485 411 7 4 89
0 3 1 151236 3056 1184 2000 148 17716 196 17716 411 360 3 8 89
0 3 1 168032 2560 1176 2008 100 15928 160 15952 443 371 4 6 91
2 1 1 187324 1784 1180 2000 164 20480 460 20480 516 421 6 7 87
2 1 1 200484 1784 1184 2004 152 14752 212 14752 475 372 1 8 91
0 3 1 220224 3432 1184 2000 228 18560 288 18560 483 437 4 8 89
2 0 2 236512 1640 1200 2000 116 16384 160 16396 437 392 5 7 88
0 3 1 253836 3188 1184 2000 84 15284 132 15316 373 343 2 5 93
1 1 1 270624 2052 1176 2000 104 19536 160 19536 508 581 4 8 88
At this point, mp3blaster is smooth again.
2 1 1 294176 1628 1176 2004 0 23420 68 23420 544 741 5 8 87
2 2 1 317728 3572 1184 2000 0 23588 60 23588 532 740 3 7 90
0 2 1 335136 3584 1180 2000 0 18080 68 18104 482 669 4 7 90
0 2 1 356128 2040 1176 2004 0 20244 68 20244 511 702 5 8 87
0 2 1 373024 1656 1184 2004 0 17960 72 17960 521 695 3 8 89
0 2 1 394528 1660 1180 2000 0 19900 300 19900 480 655 6 9 86
1 2 1 413984 2056 1184 2000 0 19924 56 19924 480 678 6 8 87
0 1 0 433952 3620 1172 2004 0 20488 60 20512 498 696 7 5 88
1 2 1 458528 1628 1176 2000 0 24100 72 24100 559 766 5 7 88
2 0 1 478496 3072 1172 1996 0 21288 68 21288 498 692 7 10 82
0 2 1 499132 1852 1176 1124 0 20172 76 20172 514 696 5 10 84
0 3 2 520632 1648 1208 1120 4 20524 72 20536 505 691 5 9 86
procs memory swap io system cpu
r b w swpd free buff cache si so bi bo in cs us sy id
0 4 1 531272 2792 1212 1164 216 11492 548 11504 574 896 3 3 94
1 3 1 539920 2144 1208 1320 52 7864 444 7864 411 624 1 6 93
0 3 1 556728 1664 1180 1188 104 18012 180 18012 518 736 5 12 82
3 0 1 573612 1784 1188 1296 32 17096 452 17096 500 698 4 9 88
1 1 1 590508 1784 1176 1212 0 16380 68 16404 473 675 4 6 90
2 1 1 610988 1912 1184 1212 0 20820 84 20820 529 723 2 11 87
1 1 1 629420 2412 1172 1216 0 18148 72 18148 474 660 2 10 88
2 1 1 650924 1744 1180 1220 0 21452 72 21452 544 786 5 9 86
0 2 1 669868 1600 1180 1216 0 19388 64 19388 468 650 3 10 88
1 2 1 688300 2552 1180 1216 0 16476 64 16512 462 646 4 11 85
0 2 1 706220 3032 1184 1216 0 19212 72 19212 487 682 7 8 86
0 2 1 728748 2548 1176 1220 0 23644 64 23644 530 712 2 15 84
2 0 0 13904 460316 1228 2292 220 6784 1364 6784 508 765 4 20 76
2 0 0 13904 196080 1244 2292 0 0 0 36 282 559 42 58 0
1 3 1 20304 2796 1184 1392 116 10740 152 10740 568 716 21 32 48
4 0 1 37960 1784 1176 1392 276 17428 348 17428 551 470 4 9 87
0 2 1 53588 2208 1176 1388 248 15076 308 15076 469 439 4 4 93
Each iteration behaves consistantly.
Scripty for summarizing mmtest output:
#!/usr/bin/perl -w
# chkmmt - check mmtest output (/usr/bin/time -v and mtest01)
# variables for interesting information
my ($allocated, $utime, $stime, $pct_cpu, $w_min, $w_secs, $maj_faults, $min_faults);
my ($tot_allocated, $tot_utime, $tot_stime, $tot_pct_cpu, $tot_w_secs,
$tot_maj_faults, $tot_min_faults);
my $count_runs;
# junk
my ($p, $d, $b, $a, $u, $t, $s);
#PASS ... 1244659712 bytes allocated.
# User time (seconds): 2.01
# System time (seconds): 2.84
# Percent of CPU this job got: 8%
# Elapsed (wall clock) time (h:mm:ss or m:ss): 0:59.02
# Major (requiring I/O) page faults: 108
# Minor (reclaiming a frame) page faults: 304655
# read a logfile from mmtest script (stdin)
while (<>) {
if (/allocated/o) {
($p, $d, $allocated, $b, $a) = split;
$tot_allocated += $allocated;
$count_runs++;
} elsif (/User/o) {
($u, $t, $p, $utime) = split;
$tot_utime += $utime;
} elsif (/System/o) {
($s, $t, $p, $stime) = split;
$tot_stime += $stime;
} elsif (/Percent/o) {
s/.* (\d+)%/$1/;
$percent = $_;
$tot_pct_cpu += $percent;
} elsif (/Elapsed/o) {
s/.*\): //o;
($w_min, $w_secs) = split ':';
$tot_w_secs += ($w_min * 60);
$tot_w_secs += $w_secs;
} elsif (/Major/o) {
s/.*: //o;
$maj_faults = $_;
$tot_maj_faults += $maj_faults;
} elsif (/Minor/o) {
s/.*: //o;
$min_faults = $_;
$tot_min_faults += $min_faults;
}
}
# print summary:
print "Averages for $count_runs mtest01 runs\n";
print "bytes allocated: ", $tot_allocated / $count_runs, "\n";
printf "User time (seconds): %.3f\n", $tot_utime / $count_runs;
printf "System time (seconds): %.3f\n", $tot_stime / $count_runs;
printf "Elapsed (wall clock) time: %.3f\n", $tot_w_secs / $count_runs;
print "Percent of CPU this job got: ", $tot_pct_cpu / $count_runs, "%\n";
print "Major (requiring I/O) page faults: ", $tot_maj_faults / $count_runs, "\n";
print "Minor (reclaiming a frame) faults: ", $tot_min_faults / $count_runs, "\n";
# end of script
--
Randy Hron