| 1 |
Kerin Millar <kerframil <at> fastmail.co.uk> writes: |
| 2 |
|
| 3 |
|
| 4 |
> A new tunable, "oom_score_adj", was added, which accepts values between |
| 5 |
> 0 and 1000. |
| 6 |
|
| 7 |
> https://github.com/torvalds/linux/commit/a63d83f#include/linux/oom.h |
| 8 |
|
| 9 |
|
| 10 |
FANTASTIC! Exactly the sort of info I'm looking for learn the pass, |
| 11 |
see what has been tried, how to configure it, and if it works/fails |
| 12 |
when and why! Absolutely wonderful link! |
| 13 |
|
| 14 |
|
| 15 |
> As mentioned there, the "oom_adj" tunable remains for reasons of |
| 16 |
> backward compatibility. Setting one will adjust the other per the |
| 17 |
> appropriate scale. |
| 18 |
|
| 19 |
That said, the mechanism seem too simple minded to succeed in anything |
| 20 |
but an extremely well monitored system. I think now the effort |
| 21 |
particularly in clustering codes, is to only have basis memory monitoring |
| 22 |
and control and leave the "fine grained" memory control needs to the |
| 23 |
clustering tools. The simple solution is there (in clustering) you just |
| 24 |
priortize jobs (codes), migrate to systems with spare resources, and bump |
| 25 |
other process to lower priority states. Also, there are (in-memory) |
| 26 |
codes like Apache-Spark, that use (RDD) Resilient Distributed Data. |
| 27 |
|
| 28 |
> It doesn't look as though Karthikesan's proposal for a cgroup based |
| 29 |
> controller was ever accepted. |
| 30 |
|
| 31 |
I think many of the old kernel ideas, accepted or not, are being |
| 32 |
"repackaged" in the clustering tools, or at least they are inspired |
| 33 |
by these codes.... |
| 34 |
|
| 35 |
Dude, YOU are the main{}. Keep the info flowing, as I'm sure lots |
| 36 |
of folks on this list are reading this ..... |
| 37 |
|
| 38 |
EXCELLENT! |
| 39 |
|
| 40 |
|
| 41 |
> --Kerin |
| 42 |
|
| 43 |
James |
Archives