XOA - local SR continous replication very slow

Hi forum,
hi Tom,

first of all, I appreciate the work you are doing by spreading the xcp-ng word in the world.

I stumbled over your youtube channel quiete fast as I firstly was tinkering with xcp-ng and it helped a lot. Thank you for this!

Now to my “problem”.

I have several xcp-ng servers running and I see on all of them that the local backups are slow as f*.

The local SR configuration is as follows:
2 * 4 TB NVMe in software raid type 1
2 * 8 TB HDDs in software raid type 1

I added the additional 8 TB local SR over the CLI.

Now if I’m doing a local backup job. Contious Repliation and Rolling Snapshot the transfer of the VMs is less than 30 MB/s. Therefoe the backup needs days to complete.

XOA is running on the same machine. I tried giving it 8 CPUs an 16 GB RAM, but this didn’t do the trick. I connected It via the internal managment interface (169.254.0.1).

In comparison I also have an external backup of the same Host running. It is a differen XOA which connects to the server and copys the VMs in a delta backup job. This is completley saturating my 1Gbit. The first full backup needed 9 hours and transfered 2.33 TiB of data, which is quite impressive speed.

I don’t understand why the local backup to local HDDs is so slow. I can’t figure out where the bottleneck is.

Any ideas?

Thank you in advance and greetings from Germany,
Dominik

Backups run though the XO instance so if it only has one interface then it can be the source of the slow down, also I think the current SMAPI process is still singe thread so the CPU speed can play a big role in the backups. The new SMAPI V3 is in beta right now which will be a much faster re-write.

Also from a storage design adding large drives to a VM is not ideal

Hi Tom,

here also thank you for your fast response.

Is there something I can do about it right now?

The sizes of the VHDs are set now, I can’t change them easily.

A queston here, what do you do in an Windows VM szenario, here it isn’t possible to splitt larger Partitions into multiple disks, like on linux with an LVM, or did I miss here something?

Thank you for your advice.
Dominik

As I covered in the storage design video I posted above, I attach the data to VM not the host often using iSCSI.

Thank you for your response,

I’ll try that. I thought the “detour” over a VM would slow everything down and ads additional overhead.

Hopefully I can try it tomorrow due its 23:00 here in Germany, I’ll give feedback if I got some results.

Thanks again, have nice sunday,
Dominik