7:44 p.m.
Hi Alex,
The regression also occurs on Haswell-EX and Ice Lake.
lkp-hsw-4ex1:
model: Haswell-EX
nr_node: 4
nr_cpu: 144
memory: 512G
bd1f77b8f24e5896 872102b4f76574cc393121c772 testcase/testparams/testbox
---------------- -------------------------- ---------------------------
%stddev change %stddev
\ | \
35518 -13% 31059
fio-basic/2M-performance-2pmem-xfs-mmap-50%-200s-randread-200G-tb-ucode=0x16-monitor=3472ca3d/lkp-hsw-4ex1
35518 -13% 31059 GEO-MEAN fio.read_iops
lkp-icl-2sp1:
model: Ice Lake
nr_node: 2
nr_cpu: 96
memory: 256G
bd1f77b8f24e5896 872102b4f76574cc393121c772 testcase/testparams/testbox
---------------- -------------------------- ---------------------------
%stddev change %stddev
\ | \
6942 -44% 3876
fio-basic/2M-performance-2pmem-xfs-mmap-50%-200s-randread-200G-tb-ucode=0x8b000015-monitor=3472ca3d/lkp-icl-2sp1
6942 -44% 3876 GEO-MEAN fio.read_iops
Best Regards,
Rong Chen
On 10/14/20 10:40 AM, Alex Shi wrote:
Hi Rong,
Consider the false sharing issue on the lock. and the qspinlock's character. I am
wondering if the problem only happens on some specific type of cpu, is this right?
Otherwise it would become a commond issue in kernel which is a huge disaster.
Thanks!
Alex
在 2020/9/25 上午9:41, kernel test robot 写道:
> Greeting,
>
> FYI, we noticed a -28.8% regression of fio.read_iops due to commit:
>
>
> commit: 872102b4f76574cc393121c77282afa62f3d9cb1 ("mm/lru: replace pgdat
lru_lock with lruvec lock")
> https://github.com/alexshi/linux.git lruv19.3
>
>
> in testcase: fio-basic
> on test machine: 96 threads Intel(R) Xeon(R) Gold 6252 CPU @ 2.10GHz with 256G
memory
> with following parameters:
>
> disk: 2pmem
> fs: xfs
> runtime: 200s
> nr_task: 50%
> time_based: tb
> rw: randread
> bs: 2M
> ioengine: mmap
> test_size: 200G
> cpufreq_governor: performance
> ucode: 0x5002f01
>
> test-description: Fio is a tool that will spawn a number of threads or processes
doing a particular type of I/O action as specified by the user.
> test-url: https://github.com/axboe/fio
>
>