Multi nic traffic output

[ranjitcool]

Hey Guys,

I have a situation here.

I want to be able to send 4gb of traffic out the nic over a vswitch port group.

Currently there are 4 1gbps nics teamed to this port group over the default - route based on virtual port id. Now I know if a vm wants to send traffic out, the traffic takes the path of the nic that was assigned to the virtual port of that vm at that instant - confirm if my understanding is correct and please correct me if i am wrong here.

Now what path policy should be chosen where the vm can send a total of 4gbps traffic using all 4 nics. Any suggestions? I am sure that will require switch port changes on the phy switch side.

Thanks

RJ

[a_p_]

The most outgoing bandwidth you can have without the NEXUS 1000V add-on is by configuring EtherChannel and using the IP-Hash policy on the vSwitch (see http://kb.vmware.com/kb1004048). However this will not give you 4GBit/s to a single destination. The VM will send outgoing traffic over the different uplinks depending on the IP-SRC-DST hash (max. 1 GBit/s to a single destination)

André

[ranjitcool]

Thanks Andre,

No we are not using any distrubited switches (if we were would this have been possible?)

If i configure etherchannel to use the ip hash policy - any changes need to be done at the physical switch port level? Basically this policy will be a faster but aggregate bw to vm will still be 1gb?

RJ

[rickardnobel]

No we are not using any distrubited switches (if we were would this have been possible?)

No, not with the VMware Distributed vSwitch.

If i configure etherchannel to use the ip hash policy - any changes need to be done at the physical switch port level?

Yes, you would have to configure the physical switch with static link aggregation, called Etherchannel on Cisco switches ("Trunk" on HP). The mode must not be "dynamic" but "on", that is there is no support for dynamic link aggregation protocol like LACP or PaGP.

Basically this policy will be a faster but aggregate bw to vm will still be 1gb?

The load balancing will be based on IP addresses. This means if you have, say, a large fileserver with hundreds of client connections then this VM could send 4 Gbit of data simulateous. However, from this VM to any single IP address there will only be one vmnic used.