what are the most stable disk settings

[RobFantini]

Hello

we are using pve5 . sometimes there are new most reliable storage settings with storage software upgrade. i/o speed tests are not a concern.

zfs and ceph , what are the most stable cache setting for storage?

with for most stable/reliable - should discard be used?

any other best practices?

thanks in advance, rob

 

[gosha]

Hi Rob!

These two storages have different purposes. They are difficult to compare with each other.

CEPH is a cluster storage (block level storage), shared between cluster nodes. Stable, reliable, regularly improves.
For reliability needs three nodes minimum. It is very desirable to use SSD for performance.
A very good choice for HA-clusters.

ZFS is a local storage (file level storage). Not shared between cluster nodes. Quite stable.
By default, it takes half the node's memory (but can be reduced).
In fact, this is like very improved software RAID.

See Table 1 in https://pve.proxmox.com/wiki/Storage

Best regards,
Gosha

 

[aderumier]

for ceph, cache=writeback can improve sequential write (but increase latency for read).

you can use discard with ceph && zfs, no problem.

 

[RobFantini]

thanks for the responses.

we use ceph and zfs,

I am looking for the most reliable cache settings for each. for data writing reliability .

 

[mir]

For zfs the safest and also given best performance is 'nocache'. This also implies the pool setting sync=standard