server load balancing (same mac address on multiple NICs)

[podera]

I believe that on UNIX operating system there is a way to configure multiple NICs with the same mac address and so realise a load balancing btw multiple interfaces. Is this something what would work in a cisco LAN infrastructure? Like two 3560s uplinked together, and both switches have one NIC with the same mac-address? (same VLAN)

[ralphcarter · Accepted solution]

Your talking about NIC bonding. You cannot have the same mac-address tied to 2 NICs at once because you will create MAC Flapping on the switches. The only way to do this is to create an LACP bond with your NIC interfaces & the switch interfaces, but these have to be either stacked switches(3750G only) or 2 x 6500 in VSS configuration.

In your case with 2 3560's, you cannot run LACP to 2 different switches, you can however run LACP to a single switch and have 2-8 NICs connecting into this 1 switch, the LACP port-channel will see 1 mac-address that is basically shared between multiple NICs on the server.

NIC Bond Modes

Mode 0 - balance-rr

Single switch/stack configuration or multiple switch,create non-lacp etherchannels on switch

This also creates a lot of “out of order” packets and forces TCP congestion control to kick in because it balances a tcp stream.

Mode 1 - active-backup

Can use on multiple switches - best for High availability

Mode 2 - balance-xor

Single switch/stack configuration, create non-lacp etherchannels on switch

Mode 3 - broadcast

Multiple switch configuration or stack that is not trunked together and connects to outside networks

Mode 4 - LACP

Single switch/stack configuration , create lacp etherchannels on switch

Mode 5 - balance-tlb (transmit load balancing)

Can use on multiple switches

Mode 6 - balance alb (adaptive load balancing, both tx and rx)

Can use on multiple switches

If you plan on connecting into 2 different 3560 switches then you can use Mode 0,1,5,6

Note: Using mode 6, the “updelay” needs to be set to something equal or greater than the switches forwarding-delay, the cisco default forwarding delay is set to 15. Cisco value of 15 is determined based upon the STP Diameter traversing 7 switches to a layer 2 destination. If you change the spanning-tree diameter then you should change the “updelay” to this forwarding-delay.

Also, you cannot have "portfast" enabled on your bonded switch interfaces, portfast can be enabled with LACP, but you cannot use this in your config.

Hope this helps.