Re: [PATCH] blk-mq: allow hardware queue to get more tag while sharing a tag set

[yukuai (C)]

On 2021/08/06 10:43, Bart Van Assche wrote:
> On 8/5/21 6:50 PM, yukuai (C) wrote:
> > After applying this configuration, the number of null_blk in my
> > machine is about 650k(330k before). Is this still too low?
>
> That seems low to me. If I run the attached script on a six year old
> desktop with an eight core i7-4790 CPU it reports a little more than 5
> million IOPS. Has kernel debugging perhaps been enabled in the kernel on
> the test setup? Or is the system perhaps slowed down by security
> mitigations?

Hi, Bart

Sorry for the delay. I was too busy with other things recently.

After disable all the kernel debuging config I can think of, the
numbers can increase to millions.

setup cmd:
modprobe null_blk nr_devices=0 &&
    udevadm settle &&
    cd /sys/kernel/config/nullb &&
    mkdir nullb0 &&
    cd nullb0 &&
    echo 0 > completion_nsec &&
    echo 512 > blocksize &&
    echo 0 > home_node &&
    echo 0 > irqmode &&
    echo 1024 > size &&
    echo 0 > memory_backed &&
    echo 2 > queue_mode &&
    echo 1 > power ||
    exit $?

test cmd:
fio -filename=/dev/nullb0 -name=test -ioengine=io_uring -direct=1
-numjobs=32 -iodepth=32 -bs=4k -rw=write -group_reporting -runtime=30
--thread --gtod_reduce=1 --ioscheduler=none -time_based

test result:
| test round | with this patch | without this patch |
| ---------- | --------------- | ------------------ |
| 1          | 4310k           | 4265k              |
| 2          | 4295k           | 4327k              |
| 3          | 4217k           | 4213k              |
| 4          | 4355k           | 4236k              |
| 5          | 4315k           | 4337k              |
| average    | 4294k           | 4275k              |

Thanks
Kuai