Increase transfer speed with TM_MAD "ssh"

Hello @pczerny @atodorov_storpool,

I have configured my datastores with TM_MAD=ssh because each host has its own disk space and there aren’t shared filesystem. However, I have noticed ssh copy (during instantiation VM time or during migration) is too slow, and it seems that with TM_MAD=qcow2 transfer would be faster.

So, would it be possible to use a non-shared datastore with ssh but increase transfer in the same way that qcow2 does? Or would it be possible to use TM_MAD=qcow2 with a non-shared datastore?

And, if I add a new network card 10 Gb, “scp” with copy (assuming instantiation VM process uses “scp” to copy VM from host to host and for migration process between hosts), will “scp” automatically take all 10 Gb or OpenNebula is limited to some speed/data_transfer?


from my experience issue is the ssh performance, that does not transfer with high speed and standard ciphers etc.
We used HPN-SSH FAQ | PSC in the past for that, but checking if new ciphers are better now.
(with the hpn patches and using “none cipher” - transfer was only internal network - we reached > 600MB/sec)
With different ciphers (to be tested) and current CPUs that should be reachable as well with a non patched ssh).

I have seen also that the latest changes in the transfer drivers (tar instead of rsync) als reduced performance, but had no time yet to take a deeper look into it (we also use shared storage so it is not that urgent for us)

Also qcow2 should not help as the bottleneck is the transfer speed.

Best, Michael