Follow-up fixes for v4.19-rc1 NUMA balancing
Get series with b4
b4 am [email protected]Versions newest first
- v1 Latest4 patches 17 repliesIn review
Patches in v1 No applies recorded yet
| # | Subject and files | Size | Replies | Applied |
|---|---|---|---|---|
| 1/4 | sched/numa: Remove redundant numa_stats nr_running fieldkernel/ | +0 | 0 replies | |
| 2/4 | sched/numa: Remove unused calculations in update_numa_statskernel/ | +1 | 0 replies | |
| 3/4 | sched/numa: Stop comparing tasks for NUMA placement after selecting an idle corekernel/ | +6 | 2 replies | |
| 4/4 | sched/numa: Do not move imbalanced load purely on the basis of an idle CPUkernel/ | +1 | 13 replies |
Reviews trailers per patch
No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.
Cover letter [PATCH 0/4] Follow-up fixes for v4.19-rc1 NUMA balancing
Srikar had an automatic NUMA balancing series merged during the 4.19 window and there some issues I missed during review that this series addresses.
Patches 1-2 are simply removing redundant code and calculations that are never used.
Patch 3 makes the observation that we can call select_idle_sibling during NUMA placement several times for each idle CPU on a socket. The patch stops the search on the first idle CPU. While there is the risk that parallel callers will stack on the same idle CPU, the current code has the same problem.
Patch 4 is the one that needs the most examination. I believe intent of load_too_imbalanced was to stop automatic NUMA balancing fighting the load balancer. Unfortunately, when a machine is lightly utilised there are idle CPUs and tasks are allowed to migrate even though the load is imbalanced. In some cases, this limits memory bandwidth and can perform worse even if data locality is fine. The patch corrects the condition.
If they pass review, I think it would suggest considering them a fix for 4.19 instead of deferring to 4.20.
kernel/sched/fair.c | 35 ++++++++--------------------------- 1 file changed, 8 insertions(+), 27 deletions(-)
Signature · 1 line
2.16.4