[zfs-discuss] Very high availability with ZFSOnLinux

Olivier Doucet webmaster at ajeux.com
Tue Apr 12 05:08:48 EDT 2016


Hello everyone,

I'm looking for very high redundancy based on ZoL. My objective is RTO < 30
seconds and RPO = 0
https://en.wikipedia.org/wiki/Recovery_time_objective
https://en.wikipedia.org/wiki/Recovery_point_objective

I have two chained SAS drives bays (with dozens of drives), hundreds of
zvol, hundreds of filesystems, and more than 10k snapshots total.
Based on my experience, the OS is the main failing part (Kernel panic,
ZFSOnLinux bugs, NFS hangs ...).
That's why I explored several solutions and kept two :

- Double controller linked to the same SAS bays.
  Downside : switching over requires a zpool import -f that takes several
minutes (even with latest zfs release that greatly improves this)

- Two mirror installations (so twice the hardware), with DRBD
active/active, multipath (ALUA). This costs much more but in theory this is
an answer to my needs.


What about you ? Tell me what you have imagined and put into production to
achieve high availability !

Olivier
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.zfsonlinux.org/pipermail/zfs-discuss/attachments/20160412/3127bb39/attachment.html>


More information about the zfs-discuss mailing list