Linux over Vmware nics show wrong mac-address to switch

[Fulanopio]

Hi. I'm new to this forum.

I wonder if this situation is common.

Tried searching for a clue everywhere but came to no success.

Our HP Blade 7000 running 4 BL460 quad nic blades is running ESX 4.1

Those nics are devoted to DMZ, LAN, INTERNET and ADMINISTRATION.

As an example, if an any windows server is setup as a vm and 2 nics are configured looking over the LAN nic (segment), lets say eth0 192.168.1.1/24 mac 00:00:00:11:11:11 and eth1 192.168.2.2/24 mac 00:00:00:22:22:22, pinging to these networks from the LAN succeeds,

Ping 192.168.1.1

Pinging........ success 4/4

Arp -a

192.168.1.1  - 00:00:00:11:11:11

Ping 192.168.2.2

Pinging........ success 4/4

Arp -a

192.168.2.2  - 00:00:00:22:22:22

Correct. But now lets setup any linux server, which is setup much the same way: the vm and 2 nics are configured looking over the LAN nic (segment), lets say eth0 192.168.1.1/24 mac 00:00:00:11:11:11 and eth1 192.168.2.2/24 mac 00:00:00:22:22:22

Ping 192.168.1.1

Pinging........ success 4/4

Arp -a

192.168.1.1  - 00:00:00:11:11:11

Ping 192.168.2.2

Pinging........ success 4/4

Arp -a

192.168.2.2  - 00:00:00:11:11:11  ¡wrong mac!

Now, changing the order of pings....

Ping 192.168.2.2

Pinging........ success 4/4

Arp -a

192.168.2.2  - 00:00:00:22:22:22

Ping 192.168.1.1

Pinging........ success 4/4

Arp -a

192.168.1.1  - 00:00:00:22:22:22 ¡wrong mac!

HP support and VmWare support couldn't provide any clue.

As can be seen, when a linux vm looks into the lan, when more than one interface is assigned, clients get their packets wrongly routed for part of their traffic (the traffic routed through the wronged interface).

Communication inter vm's is correct, and every multi nic server in the mesh does its job beautifully.

We've also dedicated a single BL460 as a stand alone test server, and tested windows and linux multi nic servers looking into the lan, and have found them to behave as supposed. So can't blame HP 7000/BL460 o Linux.

¿Is there anything wrong with vmware?

¿Any workaround?

¿Anything we might be missing or doing wrong?

Thank you in advance.

Friedrick...

[RDPetruska]

Thread moved to ESX 4 forum.

[zying]

It's more like a problem of Linux than VMware. The problem is that broadcast in 192.168.2.2/24  can reach 192.168.1.1/24 . Linux can receive ARP from 192.168.1.1/24 on the 192.168.2.2/24 interface. Setting up VLAN on the portgroups the  two vNICs bound to should solve the problem. The 2 vNICs should be two different VLAN. Perhaps need to configure the physical switch when needed.

[Fulanopio]

Stand alone hosts using same o similar configuration won't mess mac addresses, though being in the same broadcast domain.

So said, Linux works ok while not virtualised. Then Vmware messes macs.

It's vmware as the mess is seen from outside but not from inside the virtual linux.

I agree in that a couple of vlans might do the trick. But deploying vlans means parting our network in subnetworks, wich is not disireable in certain arenas. 

[rickardnobel]

Hello and welcome to the forums. Do I understand it correctly that the ping tests are being done from some physical machine outside of the VMware environment?

The multi NIC vm has two interfaces with different IP subnet, but on the same LAN?

Communication inter vm's is correct

Does this mean that if you do the ping test from another virtual machine you do not get this unexpected behavior?