I’m a longtime watcher of Tom’s videos. I was convinced to build an XCP-NG homelab server after watching his series.
I have an existing TrueNAS Scale host/server that was largely set up with the help of the Lawrence Systems vids, so I’m hoping to have similar success with XCP-NG!
Forgive me for what might be some fairly basic questions about XCP-NG.
For storage repositories (SRs), what is the best practice when picking a storage type?
How is redundancy handled? For the non-ZFS storage types, is there an ability to do striping and/or mirroring?
It seems like Vates is deprecating XOSAN v2 in favor of XOSTOR? Is XOSTOR reasonably easy to administer via CLI, if I don’t want to pay for the Vates pro license?
How would you weigh using XOSTOR vs a share on TrueNAS?
If I want to run some small VM instances that host Postgres/Mongo databases, would it be better to use local storage?
I’ve searched around for answers to these questions, but couldn’t find anything on the Vates forums, or Reddit. Hoping that this thread can serve as a resource to other XCP-NG newbies! Thanks in advance for any and all help.
If you have a separate storage server, such as a TrueNAS, NFS performs well and is thin provisioned.
For local storage XCP-ng raid is either handled by the underlying hardware or via the command line for ZFS or Linux mdadm. There is not any raid management currently exposed in XO for those.
Hyper-converged storage is nice but require fast hardware with fast connections to perform well. I have not tried managing XOSTOR from the command line.
This video explains XCP-ng storage more in depth
To better answer the database question, you could use external storage for that, I go more in depth in this video about storage design.
When you say hyper-converged storage requires “fast hardware”, do you mean with respect to disks? My build for XCP-NG is going to have a 2P EPYC setup, with plenty of DDR4 RAM, and SAS/SATA SSDs. Is there anything I need to consider beyond that?
I gave the network storage design video a watch. It makes sense that you don’t want to have a ton of storage managed by the hypervisor, but are there are any reasons why it’s a bad idea to do ZFS (for example) local storage on the XCP-NG machine, as opposed to an NFS share?
For the hyper converged systems the writes to the drives are not only going to the local drives but also going over the network to the other drives which is more likely to be what limits the speed of the system.
Local ZFS is fine, but you have to manage it from the command line in XCP-ng.
