From: Rafael Aquini raquini@redhat.com
redhat: configs: decrease CONFIG_PCP_BATCH_SCALE_MAX
In page allocator, PCP (Per-CPU pageset) is refilled and drained in batches. The batch number is scaled automatically to improve page allocation/free throughput. But a too large scale factor may end up hurting latency, like observed on https://issues.redhat.com/browse/RHEL-59526.
We're bringing this down to 0 for RHEL, to provide a smooth linear scaling factor as opposed to the default exponential scaling factor.
Signed-off-by: Rafael Aquini raquini@redhat.com
diff --git a/redhat/configs/rhel/generic/CONFIG_PCP_BATCH_SCALE_MAX b/redhat/configs/rhel/generic/CONFIG_PCP_BATCH_SCALE_MAX index blahblah..blahblah 100644 --- a/redhat/configs/rhel/generic/CONFIG_PCP_BATCH_SCALE_MAX +++ b/redhat/configs/rhel/generic/CONFIG_PCP_BATCH_SCALE_MAX @@ -1 +1 @@ -CONFIG_PCP_BATCH_SCALE_MAX=5 +CONFIG_PCP_BATCH_SCALE_MAX=0
-- https://gitlab.com/cki-project/kernel-ark/-/merge_requests/3440
From: pbrobinson on gitlab.com https://gitlab.com/cki-project/kernel-ark/-/merge_requests/3440#note_2158908...
Is there a reason we wouldn't want this on Fedora too? ATM they're both the same but for some reason weren't de-duped into common/generic:
rhel/generic/CONFIG_PCP_BATCH_SCALE_MAX:CONFIG_PCP_BATCH_SCALE_MAX=5 fedora/generic/CONFIG_PCP_BATCH_SCALE_MAX:CONFIG_PCP_BATCH_SCALE_MAX=5
From: Justin M. Forbes on gitlab.com https://gitlab.com/cki-project/kernel-ark/-/merge_requests/3440#note_2158956...
The De-dup script gets weird when things are not y, m, or not set. I can fix up Fedora though.
kernel@lists.fedoraproject.org
-
Justin M. Forbes (via Email Bridge) -
pbrobinson (via Email Bridge)
-
Rafael Aquini (via Email Bridge)