Starting with these devices (MPxIO)
I wanted to create a mirrored zpool for great data security as these two devices are located in seperate floors and rooms of the building.
Easy enough, I decided to do the simple (naive?) approach:
After some copying many files on /myvolume all looked perfect:
Looking at my devices I noticed that every 5 seconds (or so), the device stops accepting data and all drives lights are running mad for 2 seconds. Looking at the management console of these devices I saw that the whole internal RAM write buffer is written out in these intervals. Absolute performance cutter.
It seems that ZFS is perfect for hard drives which have the integriety and reliability of a politician, so it flushes very often. My RAID devices obey to these flush commands, and there is no possibility to set them to ignore these requests.
Just use this in /etc/system:
and reboot.
Transfer rates to these devices have have quadrupled since this. The setting is not critical for me, as my devices do have battery buffered RAM caches and they turn write caching immediately off if there is a problem with batteries.
1. c6t600D0230006B66680C50AB7821F0E900d0 <ADVUNI-OXYGENRAID 416F4-347B-3.41TB>
/scsi_vhci/disk@g600d0230006b66680c50ab7821f0e900
2. c6t600D0230006C1C4C0C50BE5BC9D49100d0 <ADVUNI-OXYGENRAID 416F4-347B-3.41TB>
/scsi_vhci/disk@g600d0230006c1c4c0c50be5bc9d49100I wanted to create a mirrored zpool for great data security as these two devices are located in seperate floors and rooms of the building.
Easy enough, I decided to do the simple (naive?) approach:
zpool create myvolume mirror c6t600D0230006B66680C50AB7821F0E900d0 c6t600D0230006C1C4C0C50BE5BC9D49100d0After some copying many files on /myvolume all looked perfect:
# zpool list
NAME SIZE USED AVAIL CAP HEALTH ALTROOT
myvolume 3.41T 735G 2.69T 21% ONLINE -Looking at my devices I noticed that every 5 seconds (or so), the device stops accepting data and all drives lights are running mad for 2 seconds. Looking at the management console of these devices I saw that the whole internal RAM write buffer is written out in these intervals. Absolute performance cutter.
It seems that ZFS is perfect for hard drives which have the integriety and reliability of a politician, so it flushes very often. My RAID devices obey to these flush commands, and there is no possibility to set them to ignore these requests.
Just use this in /etc/system:
set zfs:zfs_nocacheflush=1and reboot.
Transfer rates to these devices have have quadrupled since this. The setting is not critical for me, as my devices do have battery buffered RAM caches and they turn write caching immediately off if there is a problem with batteries.
Leave a comment