Proxmox with SAN

[casalicomputers]

Hi everybody,
I'm going to install proxmox on one node with a FC SAN (I would like to add another node in future), but I just have some concerns about which storage backend to use.

I really like ZFS and I would like to use it on my SAN, but I don't have RAM to give it and I'd prefer to use my SAN's RAID and caching functionality.
Is this supported somehow? (eg. create a pool with a SAN LUN only, and setting zfs arc_limit to 0 ...)
How about performance?

If not, is there an alternative to ZFS which gives me, at least, the possibility to:

• take snapshots
• have a separate volume for each VM disk
• create thin provisioned disks
• add another node

I already tried with LVM, but it lacks for thin provisioning support....

Thanks

 

[LnxBil]

In short: No

Long answer:

You need a shared disk which is currently only LVM. No other shared storage method is supported in Proxmox.
If you'd like to try an unsupported configuration: Try GFS2. It failed for me miserably, yet there are others on this forums which reported it working.

 

[fortechitsolutions]

Hi, quick footnote for what it is worth.

I've deployed a few proxmox clusters with 'classic san' storage back end as the shared storage, and LVM on top. It works, very smoothly and very reliably. But as you say, there is no thin-provisioning, which is frustrating if it is a site where people like to spin up VMs with 500gig disk and then use only 10gig of disk space within each VM for the first ~90% of lifecycle of the VM - it means lots of wasted space on the SAN.

Possible workarounds:
(a) Buy a SAN from a vendor who is not gouging you terribly, and get plenty of disk space and 'just don't worry about it' (for example my latest deploy was using http://www.ultrastor.com/index.php/es2000fs-3u16-storage-system array, and the price point is very competitive, and performance is fine, even when using 2Tb Sata disks in the array.). Clearly this option is not much use to you if you have already bought the array.
(b) deploy your SAN disk as a NAS Filer instead, ie, use decent solid "storage controller head" as an NFS filer, which mounts SAN disk as local block fibre storage, and then exports it via NFS to your proxmox cluster. Use either 10gig ether or a decent bunch of parallel gig, or do LACP or other measures to try to improve 1gig performance. Possibly if you haven't already bought the SAN disk array, you might want to strategically test or plan to buy slightly different hardware. For example I have had absurdly good experience on 'synology' storage targets for NFS with Proxmox.

End of the day, NFS performance as storage back-end for proxmox - I find it works far better than I think it should. It inherently supports thin provisioning which is dandy.

Ultimately you may have to make 'difficult decision' about deciding what is most important.
- DiskIO (ie, assumes SAN is faster:better given your deployment build scenario)
- Thin provision (ie, assumes lots of VMs will be deployed with empty/mostly empty disk)

Possibly if you are able, playing with a test NFS storage config might help you evaluate "is NFS backed storage good enough, and if yes, huzzah" ?

It will be lovely if-when LVM has support for Thin Provision (in a way that does not cause stability / performance / crash and burn outcomes). AFAIK there is something pending in the works now, but it isn't there yet (?) if I have read recent Prox4Beta announcement properly.

Tim



------------------------------------
Tim Chipman - Fortech I.T. Solutions
http://FortechITSolutions.ca