two vms on different physical hosts can't ping each other

[Shamyy]

Hello Community,

i installed NSX on two different clusters and added the two physical servers to the same VDS and same transport zone

the prob is i can't ping from vm-1 (on 192.168.4.10) to vm-2 ( on 192.168.4.40) , i.e two vms on two different physical hosts can't ping each other.

and the ping test from monitor tab failed :

and when ssh to two physical hosts and run vmkping ++netstack=vxlan 192.168.4.19 -d -s 1600 from 192.168.4.40 , i can ping normally

but when run vmkping ++netstack=vxlan 192.168.4.20 -d -s 1600 from 192.168.4.10 , i can't ping and  get message "sendto() failed (Message too long)"

so i changed the packet size to 1400 and run command again "vmkping ++netstack=vxlan 192.168.4.20 -d -s 1400" ,and it can't ping at all

although i can ping to host ip normally with 1600 packet size !

from 192.168.4.10 to 192.168.4.40

and from 192.168.4.40 to 192.168.4.10

so is there any configuration needed to enable connectivity between two virtual machines on different physical server ?

Thanks,

Shamy

[rajeevsrikant]

I believe you are facing similar issue mentioned in the below. (Assuming that your physical switches are configured for MTU 1600)

Guest VM is unable to talk to another guest VM on different ESXi host

If so it was addressed by changing the load balancing policy at distributed port groups to Route based on IP hash.

[Shamyy]

physically i connect the first interface of physical servers to physical switch ( for management)

and connect the second interface of physical servers (which i assigned to uplink1 of VDS ) to each other by cable.

it this true connection ?

[rajeevsrikant]

So you mean that you have connected both the physical servers back to back.

If so the recommended method is to connect via physical switch & setting up the MTU & the required VLANs (Trunk)

[Shamyy]

thanks rajeevsrikant ,

but what do you mean by the required vlans ? , do you mean the  vlans used when configuring vxlan.

if you mean this i used vlan 0

[rajeevsrikant]

Normally the VLANs include VLAN for VTEP, vMotion, Core Edge segment.

If you are using the same physical Nic for management segment you need to include its VLAN also.

[cnrz]

sendto() failed (Message too long) is mostly related to MTU Problem. Although it shows MTU as 1600 configured for the VTEP vmknik interfaces, is it possible to vmkping from 4.20 to 4.19 without -d and -s which would ping with fragmentation and small packet size? If this ping is successfull than MTU may also be checked further.

http://rickardnobel.se/troubleshoot-jumbo-frames-with-vmkping/

Also what is the subnet mask of the VTEP and Management interfaces? 4.19 and 4.20 is in between 4.10 and 4.40, general best practice may be to put VTEP and Management Interfaces to different Subnets although they use different TCP/IP Stacks.

Regards,