Hi Pierre,
In the past we try a similar solution than you to increase the
realiability of storages using ZFS (in our case was ubuntu). ZFS borns
to be safe, so when you write a block the system don't write other
block until this block confirms that is ok, so it slow.
In our case, we can improve this "problem" using ZIL, it's like a cache
and you can configure it on SSD disks but the performance was not really
awesome, the VMs were very slow, and Windows machines worst. After
several months the storages began to get load on cpu and we don't knew why.
Today we use 2 servers, with ext4 again, raid-10 on PERC730 controllers,
connected with DRBD to make a real time mirror between them. This works
perfect at this moment.
Regards,
El 28/4/2023 a las 05:45, Pierre Le Fevre escribió:
Hi all,
We're working on upgrading our storage solution to a proper network
attached storage.
Before this, we had an NFS share on some mounted disks on the management
server. Our new setup is a NAS running TrueNAS (zfs) with 64 GB ram and
8x8TB, 7200 rpm hard disks mounted to cloudstack over NFS.
It seems the performance is however much lower than before, resulting in
somewhat unusable virtual machines. In VMs, IOPS can be as low as <10 IOPS,
vs 100 with the management server setup.
Is there a recommended setup for primary storage to yield better
performance?
All the best
Pierre,
kthcloud
