We optimized the freezer to reduce redundant loops, and we add helper
to make code concise.
We tested the following subtree:
0
/ | \ \
A B C 1
/ | \ \
A B C 2
.....
n
/ | \
A B C
We tested is by following steps:
1. freeze 0
2. unfreeze 0
3. freeze 0
4. freeze 1
We measured the elapsed time(ns).
n=10
freeze 0 unfreeze 0 freeze 0 freeze 1
BEFORE 106179390 94016050 110423650 95063770
AFTER 96473608 91054188 94936398 93198510
n=50
freeze 0 unfreeze 0 freeze 0 freeze 1
BEFORE 109506660 105643800 105970220 96948940
AFTER 105244651 97357482 97517358 88466266
n=100
freeze 0 unfreeze 0 freeze 0 freeze 1
BEFORE 127944210 122049330 120988900 101232850
AFTER 117298106 107034146 105696895 91977833
As shown above, after optimizations, we can save elapsed time.
By freezing 0 and subsequently freezing 1, the elapsed time is consistent,
indicating that our optimizations are highly effective.
---
v2:
- open code inside the loop of cgroup_freeze instead of inline function.
- add helper to make code concise.
- remove selftest script(There are hierarchy test in test_freeze.c, I
think that is enough for this series).
Chen Ridong (3):
cgroup/freezer: Reduce redundant traversal for cgroup_freeze
cgroup/freezer: Add cgroup CGRP_FROZEN flag update helper
cgroup/freezer: Reduce redundant propagation for
cgroup_propagate_frozen
include/linux/cgroup-defs.h | 6 +-
kernel/cgroup/freezer.c | 110 ++++++++++++++++++------------------
2 files changed, 59 insertions(+), 57 deletions(-)
--
2.34.1
