Hi, all
I test blk throttle in linux.4.15-rc6, and find that,
system can't throttle IOs accurately, when io.max is set a small value.
I make a directory 'mytest' in cgroup2 and set /dev/vdc io.max to 1M/s.
Device /dev/vdc(253:32) is mounted on /ext4 with type ext4.
$ echo "253:32 wbps=1048576" > /sys/fs/cgroup/unified/mytest/io.max
$ echo $$ > /sys/fs/cgroup/unified/mytest/cgroup.procs
Test result as follow:
$ dd if=/dev/zero of=/ext4/f1 bs=1M count=10 oflag=direct
$ iostat 1 -d -h -y -p vdc
Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
vdc
3.00 0.0k 1.0M 0.0k 1.0M
Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
vdc
5.00 0.0k 1.0M 0.0k 1.0M
Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
vdc
3.00 0.0k 1.0M 0.0k 1.0M
Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
vdc
3.00 0.0k 1.0M 0.0k 1.0M
when bs = io.max, it seems OK!
$ dd if=/dev/zero of=/ext4/f1 bs=512k count=10 oflag=direct
$ iostat 1 -d -h -y -p vdc
Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
vdc
3.00 0.0k 1.0M 0.0k 1.0M
Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
vdc
2.00 0.0k 1.0M 0.0k 1.0M
Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
vdc
2.00 0.0k 1.0M 0.0k 1.0M
Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
vdc
2.00 0.0k 1.0M 0.0k 1.0M
When bs < io.max, it also seems OK!
$ dd if=/dev/zero of=/ext4/f1 bs=2M count=10 oflag=direct
$ iostat 1 -d -h -y -p vdc
Device: tps MB_read/s MB_wrtn/s MB_read MB_wrtn
vdc
4.00 0.0k 2.0M 0.0k 2.0M
Device: tps MB_read/s MB_wrtn/s MB_read MB_wrtn
vdc
0.00 0.0k 0.0k 0.0k 0.0k
Device: tps MB_read/s MB_wrtn/s MB_read MB_wrtn
vdc
6.00 0.0k 2.0M 0.0k 2.0M
Device: tps MB_read/s MB_wrtn/s MB_read MB_wrtn
vdc
0.00 0.0k 0.0k 0.0k 0.0k
However,
when bs > io.max, the max MB_wrtn/s is 2M, which is over the limit (i.e.
io.max=1M).
We also test with bs = 10M, the result also show as above.
Reason:
In my system, a bio's BIO_MAX_PAGES=512 pages (i.e. 2M).
Since dd is sequence write, each submitted bio is 2M, when bs > 2M.
blk-throttle will dispatch an integral number of bios from struct
throtl_service_queue,
resulting in MB_wrtn/s = 2M.
Is this a bug and worth to fix?
Thanks.
Yufen Yu
--
To unsubscribe from this list: send the line "unsubscribe cgroups" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
