Summary: OOM works. Takes a while to figure out what to kill, but
eventually does the right thing.
Tests: Test OOM behavior.
"mtest01 -p 80 -w" and "mmap001 -m 500000" from LTP.
Listen to mp3 on mp3blaster sampled at 128k. (usual test).
Allocate but don't write memory
-------------------------------
Allocate 101, 120% of virtual memory. No pages are written.
Nothing dies. These tests take .05 seconds. Behavior seems
appropriate for the test.
root# mtest01 -p 101
Filling up 101% of ram which is 1574421340 bytes
PASS ... 1574961152 bytes allocated.
root# mtest01 -p 120
Filling up 120% of ram which is 1879196467 bytes
PASS ... 1880096768 bytes allocated.
Allocate (but don't write to) 500% of vm. This also
takes .05 seconds. Test gives "stopped" message with
negative number of bytes allocated.
root# mtest01 -p 500
Filling up 500% of ram which is 3679666176 bytes
stopped at -1219493888 bytes
Real OOM
--------
Now create a real oom condition. Allocate and write to 101% of vm.
mtest01 process is terminated. IRC clients, vmstat, iostat keep running.
I did this test twice, and got the same behavior. You can see the
vmstat output wasn't updated much once the oom was reached and system
time went to 100%. I.E. instead of 30+ vmstat samples, only 11 printed.
This is on an Athlon 1333.
root# time mtest01 -p 101 -w
Filling up 101% of ram which is 1591169884 bytes
Terminated
real 3m11.136s
user 0m2.780s
sys 0m25.070s
vmstat 10
procs memory swap io system cpu
r b w swpd free buff cache si so bi bo in cs us sy id
0 0 0 6500 492568 2628 5284 1 271 9 604 126 118 2 2 96
0 0 0 6500 492568 2628 5284 0 0 0 0 107 76 0 0 100
1 0 0 235004 1612 1168 4516 12 18386 12 18386 274 321 11 15 73
2 0 0 552012 1608 1168 4496 4 31613 6 31613 380 529 4 14 81
5 0 0 866504 1588 1168 4496 3 31539 3 31539 381 540 5 15 80
3 2 0 1052216 1836 1172 4500 8 22490 10 22490 427 451 2 67 31
5 0 0 1052216 1668 1168 4512 44 559 48 559 201 146 0 99 0
2 0 2 1052216 1532 1208 4516 18 2 27 8 240 186 0 100 0
2 5 1 8176 497260 1172 4308 37 24 38 24 1038 800 0 100 0
0 0 0 7668 496700 1412 3992 93 0 111 21 128 91 0 1 99
0 0 0 7652 496688 1420 3992 0 0 1 0 172 193 0 0 100
Here is an oom test from Jay Thorne. I tried it on 2.4.14-pre5 and didn't
get an oom. On 2.4.14-pre5; what I saw in vmstat was at the end of the cycle,
system time went to 100%. I ended up doing an <alt sysreq Sync Term>, then
relogin on pre5.
On 2.4.14-pre6, the perl hog is terminated, and all my other processes kept
running. Yea! I ran this one twice too.
root# time nice perl -e 'while (1) { $foo .= " " x 512 } '
Terminated
real 2m1.349s
user 0m8.200s
sys 0m19.180s
vmstat 10
procs memory swap io system cpu
r b w swpd free buff cache si so bi bo in cs us sy id
1 0 0 7292 495648 1812 4004 1 293 8 620 126 117 2 2 96
0 0 0 7292 495648 1812 4004 0 0 0 0 105 76 0 0 100
1 0 0 190348 1624 1168 4244 12 13829 62 13829 235 348 29 16 55
1 0 0 460036 1624 1168 4244 0 26970 0 26970 320 441 14 15 71
2 0 0 729596 1580 1168 4244 3 26893 3 26893 325 455 14 16 70
2 1 0 998512 1608 1172 4252 9 26957 16 26956 332 481 13 17 70
5 0 1 1052216 1848 1168 4252 3 9011 6 9012 263 227 3 80 17
3 0 1 1052216 2060 1168 4252 2 841 4 841 218 174 1 98 1
2 4 0 8036 497756 1180 3976 6 2 7 2 456 375 0 100 0
0 0 0 7892 497412 1232 3860 40 0 44 5 105 69 0 0 100
Below is the results of mtest01 and mmap001 tests:
2.4.14-pre6
-----------
mp3 played 225 seconds of 327 second test.
Averages for 10 mtest01 runs
bytes allocated: 1230294220
User time (seconds): 2.089
System time (seconds): 3.013
Elapsed (wall clock) time: 32.650
Percent of CPU this job got: 15.00
Major (requiring I/O) page faults: 106.9
Minor (reclaiming a frame) faults: 301158.2
mp3 played 847 seconds of 848 second test.
Average for 5 mmap001 runs
bytes allocated: 2048000000
User time (seconds): 19.558
System time (seconds): 17.462
Elapsed (wall clock seconds) time: 169.53
Percent of CPU this job got: 21.40
Major (requiring I/O) page faults: 500163.0
Minor (reclaiming a frame) faults: 45.0
Portions of the two previous kernels:
2.4.14-pre5
-----------
mp3 played 263 seconds of 321 second run.
Averages for 10 mtest01 runs
bytes allocated: 1241933414
Elapsed (wall clock) time: 32.104
Percent of CPU this job got: 15.70
No mp3 skips noted.
Average for 5 mmap001 runs
bytes allocated: 2048000000
Elapsed (wall clock seconds) time: 173.48
Percent of CPU this job got: 20.80
2.4.14-pre4 (2 oops)
-----------
mp3 played 250 of 313 seconds.
Averages for 10 mtest01 runs
bytes allocated: 1244554854
Elapsed (wall clock) time: 31.306
Percent of CPU this job got: 16.00
mp3 played 860 of 865 seconds.
Average for 5 mmap001 runs
bytes allocated: 2048000000
Elapsed (wall clock seconds) time: 172.63
Percent of CPU this job got: 21.60
Have fun!
-- 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/