gns3 cluster at esxi bottleneck

[Sardaukai]

Picture for illustration

Hi all,
I made some benchmark tests with our gns3vm servers (2.2.40.1). It was really good, but I found one bottleneck.
I used iperf3 in order to get a measurable throughput.

Linux_A to Debian
PASS: ping
PASS: iperf3 ~2Gbit/s

Cluster, working, but slow:
Linux_A is connected directly via eth1 with GNS_Cloud_A
Linux_B is connected directly via eth1 with GNS_Cloud_B

PASS: ping from Linux_A to Linux_B
PASS: iperf3 ~50 Mbit/s

Cluster with bottleneck:
Linux_A is connected via br1 with GNS_Cloud_A
Linux_B is connected via br1 with GNS_Cloud_B

FAIL: ping

Error:
prePing - arp request, is sent from Linux_A to Linux_B
I am able to see the reply with tcpdump at interfaces A_br1 and A_eth1 but not at gns3tap0-0
or with wireshark at cable to Linux_A

Workaround:
I set static arp addresses on both sides.

PASS: ping
PASS: iperf3 about 1 Gbit/s

Question:
Is there a way to configure the vm without that workaround?
I monitored iptables, arptables, ebtables and different driver settings without any hint of a drop.
Unicasts from Debian working fine, but every unicast frame from Linux_B to Linux_A
hits A_br1 and A_eth1 but not gns3tap0-0 or at cable to Linux_A.
It is simply gone ;/

[Sardaukai]

Just another workaround:
UDP tunnels at the cloud device instead of ethX or brX, works as point-to-point connection, got about 1 Gbit/s again 😄

[elico]

@Sardaukai Can you share a little more details?
I am not sure I understood the setup you have in the illustration.
I have tried to connect my GNS3 VM guests to the LAN using a Cloud node and found a way to make it work very very fast.
It works as expected, please take a look at:
https://www.gns3.com/slow-speeds-on-cloud-node-anyone-else