putting up a ZFS storage
I'm putting up a ZFS storage on an old R720 Server with the PERC firmware reconfigured to allow direct disk access. this machine has 8 1.8TB sata disks. I'm gonna use it to serve disks for KVM and XCP-NG Vms. What would be the best pool setup for this old bunch? I was thinking about mirroring it in pairs, but I don't know if this would be smart at all.
14 replies
Depends what you want. Do you want to max capacity? Max IOPs? Max redundancy? Some combination of those?
They are small disks so you could probably get away with a RAIDz2 since any resilver would be fast
I'm going to try it with 2 raidz1 vdevs, as a compromise between a good amount of storage, security and performance for the VMs. My current demand for space is only 3.4TB.
Yeah someone interested in zfs! I'm all for it.
Having had a few (smaller) setups with 4 disks privately but bigger setups commercially (24) I'd simply recommend raidz#, parity amount which you'd like to have, for 8 disks 2 sounds fine.
You could do it differently but don't overthink it. This way you can lose up to # disks no matter which. And the performance should be fine but you can test that if you wish.
thanks for the answer. I'm thinking about doing two vdevs on raidz1, each one with 4 disks and build up from there.
If you need/want to split up storage, that's a good way to do so. Otherwise, it's more flexible with 8 in z2
Nope. No 'a'. It's like saying 'a food'.
acknowledge, my bad :D
What's your backup strategy? How precious the data is could influence your pool configuration.
Well, it would be really bad to lose a VM, for sure. For the time being, this vms will be backed on a bareos on other phisical server.
Since you've got backups your needs for redundancy are low, could just z1 it or pool of mirrors.
Well, Im trying it with the 2 raidz of 4 disks. but now, other questions emerged
NFS or iSCSI Considering that my enviroment is composed by XCP-NG hosts and a ZFS pool of 2 vdevs with 4 disks, what would be the pros and cons of chossing NFS (file-based) or iSCSI/Zvol (block storage SR)? I'm concerned about the high density of VDIs and IOPS on a sync writes. The server has an adequate amount of ram (64GB)
mitigate the sync writes. XCP-ng, one of my vm manager (other is going to be KVM), by default, usually force sync writes to garatee VM integrity. That could be bad for the performance of this SAS 7200RPM disks. I don't have, at least for now, a good solid state drive to help, but maybe some tuning on the sync on the ZFS could help? (sync=always vs sync=standard vs sync=disabled)
ZFS volume tuning I set up my ZFS pool with ashift=12 (4K) and recordsize=128k. If i choose for iSCSI creating ZVOL volumes, what should be the ideal volblocksize?
Backup and snapshot Right now I have a pretty big archived old VMs and snapshots. I wonder what would be more efficient for backup and recovery. Use zfs snapshot and tools for backup or use the orchestrator on XCP-NG?
honestly if your running linux i'd just go with bcachefs these days.
what would be the advantages?
better performance (its getting close to EXT4/XFS and on par in some situations), better durability, easier to setup and maintain, tiered storaged (ssd's front hdds).