diff mbox series

Reduce ksm06 memory footprint

Message ID 20210119134057.17785-1-mdoucha@suse.cz
State Accepted
Headers show
Series Reduce ksm06 memory footprint | expand

Commit Message

Martin Doucha Jan. 19, 2021, 1:40 p.m. UTC
Allocating 10,000 pages requires over 600MB on PPC64. This may result in some
pages being pushed to swap even if the test machine has 1GB of RAM per NUMA
node. Reduce memory footprint a little to avoid bogus failures.

Signed-off-by: Martin Doucha <mdoucha@suse.cz>
---
 runtest/mm | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Cyril Hrubis Jan. 19, 2021, 3:23 p.m. UTC | #1
Hi!
> -ksm06_2 ksm06 -n 10000
> +ksm06_2 ksm06 -n 8000

Applied, since this is a sensible fix for the release.

Ideally we should change the test to be able to scale with the available
memory later on.

I guess that we can easily add array of free memory per node for the
tst_numa.h nodeamp, then we can select subset of nodes based on
allocation size or allocating size based on the set of nodes.
diff mbox series

Patch

diff --git a/runtest/mm b/runtest/mm
index 481d39691..eeebd878e 100644
--- a/runtest/mm
+++ b/runtest/mm
@@ -70,7 +70,7 @@  ksm04_1 ksm04 -u 128
 ksm05 ksm05 -I 10
 ksm06 ksm06
 ksm06_1 ksm06 -n 10
-ksm06_2 ksm06 -n 10000
+ksm06_2 ksm06 -n 8000
 
 cpuset01 cpuset01