| 1 |
On 18/09/2014 19:27, James wrote: |
| 2 |
> Kerin Millar <kerframil <at> fastmail.co.uk> writes: |
| 3 |
> |
| 4 |
> |
| 5 |
>> The need for the OOM killer stems from the fact that memory can be |
| 6 |
>> overcommitted. These articles may prove informative: |
| 7 |
> |
| 8 |
>> http://lwn.net/Articles/317814/ |
| 9 |
> |
| 10 |
> Yea I saw this article. Its dated February 4, 2009. How much has |
| 11 |
> changed with the kernel/configs/userspace mechanism? Nothing, everything? |
| 12 |
|
| 13 |
A new tunable, "oom_score_adj", was added, which accepts values between |
| 14 |
0 and 1000. |
| 15 |
|
| 16 |
https://github.com/torvalds/linux/commit/a63d83f#include/linux/oom.h |
| 17 |
|
| 18 |
As mentioned there, the "oom_adj" tunable remains for reasons of |
| 19 |
backward compatibility. Setting one will adjust the other per the |
| 20 |
appropriate scale. |
| 21 |
|
| 22 |
It doesn't look as though Karthikesan's proposal for a cgroup based |
| 23 |
controller was ever accepted. |
| 24 |
|
| 25 |
--Kerin |
Archives