Messages in this thread | | | Subject | Re: [RFC PATCH v8] sched/fair: select idle cpu from idle cpumask for task wakeup | From | "Li, Aubrey" <> | Date | Tue, 15 Dec 2020 20:41:29 +0800 |
| |
Hi Bao Hua,
Sorry I almost missed this message, :(
On 2020/12/14 7:29, Song Bao Hua (Barry Song) wrote: > > Hi Aubrey, > > The patch looks great. But I didn't find any hackbench improvement > on kunpeng 920 which has 24 cores for each llc span. Llc span is also > one numa node. The topology is like: > # numactl --hardware > available: 4 nodes (0-3) > node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 > node 0 size: 128669 MB > node 0 free: 126995 MB > node 1 cpus: 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 > 43 44 45 46 47 > node 1 size: 128997 MB > node 1 free: 127539 MB > node 2 cpus: 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 > 67 68 69 70 71 > node 2 size: 129021 MB > node 2 free: 127106 MB > node 3 cpus: 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 > 91 92 93 94 95 > node 3 size: 127993 MB > node 3 free: 126739 MB > node distances: > node 0 1 2 3 > 0: 10 12 20 22 > 1: 12 10 22 24 > 2: 20 22 10 12 > 3: 22 24 12 10 > > Benchmark command: > numactl -N 0-1 hackbench -p -T -l 20000 -g $1 > > for each g, I ran 10 times to get the average time. And I tested > g from 1 to 10. > > g 1 2 3 4 5 6 7 8 9 10 > w/o 1.4733 1.5992 1.9353 2.1563 2.8448 3.3305 3.9616 4.4870 5.0786 5.6983 > w/ 1.4709 1.6152 1.9474 2.1512 2.8298 3.2998 3.9472 4.4803 5.0462 5.6505 > > Is it because the core number is small in llc span in my test?
I guess it is with SIS_PROP, when the system is very busy that idle cpu scan loop is throttled by nr(4). The patch actually reduces 4 times scan so the data change looks marginal. Vincent mentioned a notable change at here:
https://lkml.org/lkml/2020/12/14/109
Maybe you can increase the group number to see if it can be reproduced on your side.
Thanks, -Aubrey
| |