1 |
Keith M Wesolowski writes: |
2 |
in response to my: |
3 |
> |
4 |
> > make -j all |
5 |
> > |
6 |
> > uptime: 09:41:41 up 10m, 2 users, load avg: 200.90, 65.79, 23.42 |
7 |
> > ... |
8 |
> > Dec 3 09:48:05 VM: killing process apache2 |
9 |
> |
10 |
> > OOMkiller 'feels' more aggressive in -rc2, but still |
11 |
> > doesn't have the instability that gds-267-r16 has |
12 |
> |
13 |
> [...] |
14 |
> |
15 |
> I question the use of make -j as a stress test. What is the desired |
16 |
> behaviour of running this command, other than not to crash the OS? |
17 |
|
18 |
|
19 |
The desired behaviour was to hard-stress the system and determine at what |
20 |
patch-point the crashing behaviour was introduced. |
21 |
|
22 |
I reported that gds-267-r16 hard-crashed in under 2m of this stress test |
23 |
due to bugs in the memory allocation process (to my untrained eye) |
24 |
|
25 |
Jeremy asked me to check 2.6.7-rc1 and 2.6.7-rc2 to see if the bug was |
26 |
evident in either of those. |
27 |
|
28 |
I reported that it wasn't. |
29 |
|
30 |
The entire point of this is to determine why lockups are happening |
31 |
on SMP Ultra-2 sparcs. |
32 |
|
33 |
Since I'm not seeing any lockups -- except for gds-267-r16, under this |
34 |
stress |
35 |
test, this is the only method I have at this point to evaluate different |
36 |
kernels. |
37 |
|
38 |
|
39 |
My reporting of when the OOM and VM killers strike is merely to show the |
40 |
slight |
41 |
difference in kernel operation, and to show where (if any) instability might |
42 |
be creeping in. |
43 |
|
44 |
For instance, in the same period of time, the VM killer did not strike in |
45 |
-rc1, |
46 |
yet did strike in -rc2. Is this a cause for concern? Not for me. Maybe |
47 |
for |
48 |
somebody who understands the internal differences. Hence I pass on the |
49 |
info. |
50 |
Note that I was not extremely detailed in my reporting of it, although I do |
51 |
have the full logfile for anybody who is that interested. |
52 |
|
53 |
-- |
54 |
leif |