=== mamarley_ is now known as mamarley | ||
=== v12aml_ is now known as v12aml | ||
amitk | cking: hiya! How do I prevent stress-ng from running any memory stress tests (since I only have a Gig) but run everything else? | 10:09 |
---|---|---|
amitk | cking: I tried 'stress-ng -r 4 -x vm --minimize' but I still get several OOMs | 10:09 |
cking | amitk, the only easy way is to exclude all the ones that OOM by the -x option, so that's a a pain | 10:11 |
cking | however, there are quite a few stressors that can trigger OOMs. However, the OOMs won't affect stress-ng, it can respawn the stressors if they get OOM'd | 10:13 |
cking | one can specify per-stressor memory sizes in terms of memory size too (in bytes or % memory available) | 10:13 |
amitk | cking: aah, ok. The list of stressors is quite long and there seems to be no easy way to eliminate an entire class of stressors | 10:14 |
cking | amitk, yep, that is a currently a pain point | 10:14 |
amitk | cking: doesn't the --minimize flag avoid OOMs be being conservative? | 10:15 |
amitk | *by | 10:15 |
cking | amitk, --minimize just selects the lower bounds settings | 10:15 |
amitk | that is the opposite of what I thought it does :-) | 10:15 |
amitk | I guess I'm looking for max bounds on my memory and IO constrained systems | 10:16 |
cking | but this is a per-test specific setting; some stressors just require a few tens of MB to run effectively | 10:16 |
cking | and that's the lower bounds | 10:16 |
cking | amitk, i suggest running each stressor one by one with a specific memory bound set for the stressors you are interested in | 10:17 |
cking | e.g stress-ng --vm 0 --vm-bytes 80% | 10:17 |
amitk | cking: I really want random to catch unexpected errors but was hoping to find a way to say "don't commit beyond 60% of memory" | 10:18 |
cking | amitk, OK, I think one needs to just select the subset of stressors you are interested in and specify the memory size according to your desired constraints on a per-stressor basis | 10:19 |
cking | yep, it's a pain | 10:19 |
amitk | cking: ok, thanks. Its been very useful so far to test cpufreq and thermal :-) | 10:21 |
cking | amitk, that's good to know. it can make CPUs run quite hot :-) | 10:21 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!