Linux bridge on SFP+ - strange behavior

[m3a2r1]

I've got such scenario to analyze, I don't understand what I've done wrong, maybe it's a bug.

My configuration of network in PVE:
eno1 - ethernet interface (1Gbit)
vmbr0 - main bridge connected to eno1 with ip address
ens2f0 - sfp+ interface (10Gbit)
ens2f0.10 - vlan 10 on interface ens2f0
vmbr010 - bridge connected to ens2f0.10
vmbr1 - bridge connected to ens2f0

There is vm with Debian where I test communication with other Debian VM installed on PVE on another physical machine (connected to the same LAN with SFP+)
I'm testing with iperf, second Debian is a server.
So I've got such results:
vm connected to vmbr0 (ethernet) - 645Mbit (it's ethernet so it is below 1Gbit, it's ok)
vm connected to vmbr010 - 5.5Gbit (it's sfp+ so it's great)
vm connected to vmbr1 (sfp+ interface) - and there is a suprise - bandwidth is only 615Mbit and it looks like ethernet connection.

I don't have any idea what happened. Anybody can explain such behavior? Maybe I misconfigured something?

 

[spirit]

can you send the ouput of "brctl show" when it's slow, to be sure of bridge current running configuration ?

are you sure that when you test with iperf on vmbr1, the traffic is not going through an intermedate gateway ? (client && server are both on the same lan && ip subnet ?)

 

[m3a2r1]

are you sure that when you test with iperf on vmbr1, the traffic is not going through an intermedate gateway ? (client && server are both on the same lan && ip subnet ?)

I think it's the point. Communication is over sfp, but that "bad" subnet is non-vlan class other than vlan on iperf server and routing goes by router which is not connected by sfp yet (Client is on 172.16.1.0/24 and server on 172.16.0.0/24)