0

If I am running VMWare server on a Linux host, and the VMs are using bridged network, will the VMs know to failover to the other NIC? I don't have a clear picture of how VMWare bridged networking actually works.

I still need to go read more about Linux bonding. My goal is to have a secondary NIC that is failover where the switches do not need to be stacked (Each nic will be connected to a different switch, the switches will be trunked). So maybe I will ask this in a second question, and it is kind of a bonus question here :-)

Kyle Brandt
  • 83,619
  • 74
  • 305
  • 448

1 Answers1

2

Works perfectly with KVM:
ethX+ethY -> bondX -> bridgeX -> VM
failover happens on the bond level, transparent to the VM

dyasny
  • 18,802
  • 6
  • 49
  • 64
  • He was asking about VMWare Server. – pauska Feb 16 '10 at 15:11
  • shouldn't be different, just need to use the bond alias instead of the NICs when configuring the bridge there. – dyasny Feb 16 '10 at 15:49
  • This is how I have my VMWare Server setups configured but honestly I haven't yet tested they behave like we want. – Keith Stokes Feb 16 '10 at 15:55
  • So I create the bond interface, then re-run configure.pl (whatever that was) and tell it to bridge to the bondX interface (ie bond0) ? – Kyle Brandt Feb 16 '10 at 15:57
  • Not too sure about the vmware bits, I've kept away fir over a year now. but yeah, either via a script, or connect to the GUI, and set up the bridge (or v-switch?) to connect to the bondX interface as the physical NIC it should go out through – dyasny Feb 16 '10 at 17:05
  • Yes. I decided today was the day to test the whole bonding thing since I have another customer who wants to run it and it didn't work our right on our first test the other night. I just tested a CentOS 4.8 host running 1.09 or 1.10 and dual NICs. Dropping one port at the switch level made the host switch just fine and might not have even dropped a ping. The guests were not happy. I suspect it might be that they are running on shared storage via NFS and that's what freaked out. I had to acknowledge a read error in the VMWare console on most guests to make them happy. – Keith Stokes Feb 16 '10 at 17:05
  • Testing a non-shared-storage system now. – Keith Stokes Feb 16 '10 at 17:06
  • After a couple of sidetracks, here's what I found: Host: CentOS 4.8 with older kernel 2.6.9-34 and VMware Server 1.10 using local storage. Dual NICs going to dual switches and running bonding. Guests: Win2k3 Web and CentOS 5.4/Apache Running pings to all 3 machines. If I down the primary host interface at the switch, the host drops a packet and switches to the secondary. Both guests drop packets and do not return until they originate outbound traffic, either by themselves or if I log into the console. Not exactly what we want... – Keith Stokes Feb 16 '10 at 18:56
  • Rather strange... Can you try the same with virtualbox? This might be specific to the way VMWare interacts with the network. I've already mentioned KVM, but I don't know if your hardware can support it – dyasny Feb 16 '10 at 20:15
  • Hi Kieth, can you post your bonding config as an example for me? I will run some tests as well. – Kyle Brandt Feb 17 '10 at 15:53