PBS and iSCSI 10gb

[pownwen]

Hello everybody!
At the moment I have 3 servers with PVE, to which iscsi storage 10Gbit/sec is connected via multipath. On one of the nodes I installed PBS as a virtual machine, accordingly its virtual disk is located on iscsi storage. When measuring the speed of restoring virtual machines with PBS, an approximate speed of 1Gbit/sec was noticed, and there is a strong feeling that the process of restoring the virtual machine is carried out over the local network, and not directly via iscsi. Please advise whether this connection is correct or whether the storage needs to be mounted to PBS in some other way?

 

[tcabernoch]

I dunno man. I'm testing a similar setup at this very moment and getting cr@p results.

Let's compare some notes.
I followed this https://pve.proxmox.com/wiki/Multipath

In PVE cluster Storage, I did the LVM on top of a LUN mount they describe.
I have 4 paths going. 10gb network.

On PVE host, made a huge virtual disk for PBS VM on the LVM on LUN.

And then in PBS, made a ZFS datastore on that using the GUI.
I think i should add _netdev to the mount. Haven't figured out how, as its ZFS.

The VM is locking up. I've clearly done something wrong.
I'm going to dive back into this right now.

I have a bunch of iSCSI-only storage and no other use for it.
I want this to work.

...

I should add that I have another test system with iSCSI mounted in the PBS OS, and that works reasonably well.
Its a pain to setup tho, difficult to recover, just not optimal.
It seems more natural and supported to mount the iSCSI as a PVE datastore.

 

[tcabernoch]

Ok, i had a MAC ID conflict. I fixed that. It works now.

This is a shot of syncing onto the storage from the old PBS, so my bandwidth cap here might actually be the source.
Ya, this is disk write. Jumps between 200 M and 150 M.
I think that's fairly good for an array of spinners. Good enough for me, anyway.

Unless anybody else has any pointers, I'm done here.
I have PBS hosted on PVE with an iSCSI data disk, and its working well.