Hardware:
AMD Athlon 1333
512 MB ram
1024 MB swap
Notes:
Interactive response was noticeably better with 2.4.12-ac3+vmpatch.
2.4.12-ac3+vmpatch was tested right after a reboot.
2.4.12-ac3+vmpatch (page-cluster=4)
Averages for 5 mmap001 runs
bytes allocated: 2048000000
User time (seconds): 19.322
System time (seconds): 16.576
Elapsed (wall clock seconds) time: 152.87
Percent of CPU this job got: 22.80
Major (requiring I/O) page faults: 500175.2
Minor (reclaiming a frame) faults: 20.0
This is the tail end of vmstat 1 on 2.4.12-ac3+vmpatch:
procs memory swap io system cpu
r b w swpd free buff cache si so bi bo in cs us sy id
4 0 1 9280 3064 2012 481536 0 0 168 16188 438 448 0 26 74
4 0 1 9280 3064 2020 481520 0 0 136 16372 512 447 1 21 78
0 2 1 9280 3064 2036 481480 0 0 284 12148 461 465 0 20 80
5 2 1 9280 480788 2524 5124 0 0 632 11396 451 421 0 42 58
2 0 0 9280 479316 2592 6440 60 0 1872 140 671 935 0 9 91
2.4.13-pre3aa1 was tested after running 6 runalltest.sh LTP suites,
as well as normal use. uptime 40 hours.
2.4.13-pre3aa1 page-cluster=2
Averages for 5 mmap001 runs
bytes allocated: 2048000000
User time (seconds): 19.590
System time (seconds): 14.846
Elapsed (wall clock seconds) time: 213.90
Percent of CPU this job got: 15.60
Major (requiring I/O) page faults: 500171.2
Minor (reclaiming a frame) faults: 29.8
2.4.13-pre3aa1 page-cluster=4
Averages for 5 mmap001 runs
bytes allocated: 2048000000
User time (seconds): 19.220
System time (seconds): 15.378
Elapsed (wall clock seconds) time: 209.41
Percent of CPU this job got: 16.00
Major (requiring I/O) page faults: 500155.6
Minor (reclaiming a frame) faults: 31.2
2.4.13-pre3aa1 page-cluster=6
Averages for 5 mmap001 runs
bytes allocated: 2048000000
User time (seconds): 20.368
System time (seconds): 16.218
Elapsed (wall clock seconds) time: 206.08
Percent of CPU this job got: 17.20
Major (requiring I/O) page faults: 500166.6
Minor (reclaiming a frame) faults: 28.8
Just to make sure the previous tests didn't skew the results,
I rebooted into 2.4.13-pre3aa1 and ran with the default
page-cluster value:
2.4.13-pre3aa1 page-cluster=6 (after reboot)
Averages for 5 mmap001 runs
bytes allocated: 2048000000
User time (seconds): 20.364
System time (seconds): 15.798
Elapsed (wall clock seconds) time: 204.09
Percent of CPU this job got: 17.40
Major (requiring I/O) page faults: 500165.6
Minor (reclaiming a frame) faults: 30.0
tail end of vmstat from final iteration on 2.4.13-pre3aa1:
procs memory swap io system cpu
r b w swpd free buff cache si so bi bo in cs us sy id
2 1 1 8644 3372 1808 484988 248 0 324 32980 873 671 0 22 77
2 3 1 8616 3316 1824 485000 32 0 60 16192 467 405 1 27 72
1 4 1 8600 3100 1848 485012 88 0 124 12148 422 427 1 22 77
1 1 0 8532 488648 1508 1680 76 0 600 7108 439 561 4 28 68
1 1 0 8340 485796 3384 2548 240 0 2984 0 858 1484 0 41 59
This test does use a little swap with both kernels. That puzzled me,
since I thought 50000 pages was about 195 megs of RAM.
page-cluster=6 makes mp3blaster skip more on this test.
The "free -mt" output below is after the test on 2.4.13pre3aa1 to give an idea of
what memory usage is like when there is no test running.
total used free shared buffers cached
Mem: 502 32 470 0 4 6
-/+ buffers/cache: 21 480
Swap: 1027 7 1019
Total: 1530 39 1490
Actually, kernel hackers can probably tell that from the vmstat snippets. :)
-- Randy Hron- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/