Slow disk IO on OVH dedicated server

[czmiho]

I have ProxmoxVE on an OVH dedicated server using their install. They use md by default. Throughput is ok but FSYNCS/SECOND is only ~10.

Disks are 7200rpm: HGST HUS724020ALA640

pveperf /var/lib/vz
CPU BOGOMIPS: 54276.16
REGEX/SECOND: 1829496
HD SIZE: 1809.50 GB (/dev/mapper/pve-data)
BUFFERED READS: 108.65 MB/sec
AVERAGE SEEK TIME: 25.34 ms
FSYNCS/SECOND: 9.57 (!!!!!!!!)


cat /proc/mdstat
Personalities : [raid1]
md4 : active raid1 sda4[0] sdb4[1]
1931980736 blocks [2/2] [UU]
bitmap: 8/15 pages [32KB], 65536KB chunk
md2 : active raid1 sda2[0] sdb2[1]
20478912 blocks [2/2] [UU]

hdparm -W /dev/sda
/dev/sda:
write-caching = 1 (on)

hdparm -W /dev/sdb
/dev/sdb:
write-caching = 1 (on)

pve-manager/4.2-4/2660193c (running kernel: 4.4.8-1-pve)


What the heck is wrong with this ;-)

 

[dietmar]

What the heck is wrong with this ;-)

The disk is simply that slow (cheap disk)?

 

[czmiho]

The disk is simply that slow (cheap disk)?

Those are actualy quite good: http://www.storagereview.com/hgst_ultrastar_7k4000_review_hus724040ale640

Should do 150 IOPS read and 200 IOPS write

 

[elkato]

Hey czmiho,

A dedicated on OVH here too. With 3x sames disks:

# pveperf /mnt/storage2
CPU BOGOMIPS: 57597.12
REGEX/SECOND: 1766622
HD SIZE: 1833.78 GB (/dev/mapper/vg--storage2-lv--storage2)
BUFFERED READS: 156.20 MB/sec
AVERAGE SEEK TIME: 11.50 ms
FSYNCS/SECOND: 990.62

pveperf /var/lib/vz
CPU BOGOMIPS: 57597.12
REGEX/SECOND: 1662954
HD SIZE: 1760.99 GB (rpool/ROOT/pve-1)
FSYNCS/SECOND: 44.80

/mnt/storage2 a disk alone with lvm
/var/lib/vz over 2 disks in zfs raid-1

Do you have many VMs running while measurement?

 

[czmiho]

after reboot (all LXC stopped):

BUFFERED READS: 127.81 MB/sec
AVERAGE SEEK TIME: 13.64 ms
FSYNCS/SECOND: 29.20

This is still less than 1/10 of expected values...

 

[spirit]

Those are actualy quite good: http://www.storagereview.com/hgst_ultrastar_7k4000_review_hus724040ale640

Should do 150 IOPS read and 200 IOPS write

don't forget that with zfs, you write data twice (1 in journal, 1 in datas). so, for write, you can already divide by 2.

also theses numbers are when you are doing 100% read or 100%write, with 16 threads.
Don't expect to reach theses numbers in real life.

 

[czmiho]

Just quick note: FS is EXT3

 

[czmiho]

smartctl -g all /dev/sda
smartctl 6.6 2016-05-31 r4324 [x86_64-linux-4.4.8-1-pve] (local build)
Copyright (C) 2002-16, Bruce Allen, Christian Franke, www.smartmontools.org

AAM feature is: Unavailable
APM feature is: Disabled
Rd look-ahead is: Enabled
Write cache is: Enabled
ATA Security is: Disabled, frozen [SEC2]


smartctl -g all /dev/sdb
smartctl 6.6 2016-05-31 r4324 [x86_64-linux-4.4.8-1-pve] (local build)
Copyright (C) 2002-16, Bruce Allen, Christian Franke, www.smartmontools.org

AAM feature is: Unavailable
APM feature is: Disabled
Rd look-ahead is: Enabled
Write cache is: Enabled
ATA Security is: Disabled, NOT FROZEN [SEC1]