ltp icon indicating copy to clipboard operation
ltp copied to clipboard

Published 20 hours ago • verified publisher icon linux-test-project

msgstress tests should scale number of processes accodingly to available RAM

Open metan-ucw opened this issue 6 years ago • 6 comments

The msgstress tests (previously msgctl08 - msgctl11) should scale number of processes accordingly to the amount of free RAM, currently the hardcoded values needs slightly more than 1GB on x86_64 and possibly more on ppc64le due to larger page size.

metan-ucw avatar Mar 14 '19 10:03 metan-ucw

Recently, I run msgstress04 and it fails because systemd(after v243-rc1) has increased pid_max to 4194304 on 64 bit machine since sysctl: let's by default increase the numeric PID range from 2^16 to. We should drop hardcoded values.

xuyang0410 avatar May 08 '20 06:05 xuyang0410

Also, it looks like we can drop COLDFIRE like uclinux. Is it right?

xuyang0410 avatar May 08 '20 07:05 xuyang0410

We are phasing out uclinux support, so yes we can drop it. Also these tests should be converted to use the new library as well.

metan-ucw avatar May 11 '20 08:05 metan-ucw

Ok. I will convert these cases into new library(in my todo list).

xuyang0410 avatar May 12 '20 01:05 xuyang0410

@metan-ucw I am working on it. But I have a question about this. IMO, the message queue consume memory size is about MSGMNB * MSGMNI (On my system, it is 16k * 32k = 512M , is it right?) Why ppc64le or arrch64 consume more memory because of large page size(is it 64k * 32k )? Does MSGMNB need to align with page size?

xuyang0410 avatar May 29 '20 10:05 xuyang0410

This still happens. Any update? I see a patch trying to fix it, but it didn't land in repo. https://patchwork.ozlabs.org/project/ltp/patch/[email protected]/

realhezhe avatar Sep 07 '21 03:09 realhezhe