VC Flex 10 Question on Multiple Shared Uplink Sets

[chuckk281]

Donald had an interesting customer question:

******************************************************************************************************

My customer currently plans on implementing two enclosures with the single shared uplink design from the cookbook.  They have posed the question as to the effect of adding an additional uplink set as described in Scenario 3:2 which provides active/active paths for the traffic.  Would the end server nodes need additional IP addresses to take advantage of this design and if so is there a way to implement this as an active/standby design so these additional IP addresses are not required.

************************************************************************************************************

Paul gave some insight:

****************************************************************************************************************

Neither an active/passive nor active/active VC config requires more than 1 IP address on the server.

In either case, you select two NICs that are routed thru two different IO slots.  You create a team from them, and the team shares a single IP address.

In an active/active scenario, you create two different VC networks, with each network using uplink ports from a single VC module.  Both of those networks could be used simulataneously, but if the NIC teaming is set to active/passive or failover, only one NIC will receive and transmit, so traffic will be limited to the uplinks from the IO slot that NIC is routed to thru the signal midplane.

In an active/passive config, a single VC  network is created using port(s) from two different VC modules.  VC will use the port(s) from one module or the other, but not both simultaneously.

In active/active, an upstream switch or uplink failure is communicated all the way to the server OS/hypervisor, and the traffic is sent thru the other NIC in the team instead.  In active/passive, VC would just route the traffic thru the other module via interslot stacking links, and use the uplinks from the other module.  The OS/hypervisor would not need to know about the upstream switch failure, the failover is handled at the VC level.

****************************************************************************************************************

Have you implemented this scenario and how if so, how did it function for you?

[gnijs444]

True. In the Active/Active scenario, you will need Smartlink feature to physically bring down the NIC (maybe even special NIC drivers to use DCC to bring down a "logical" interface when running trunks on a physical flex-10 nic).

In Active/Passive scenario, you don't need this. A physical upstream failure will just transition the "standby" link to "active". The server NIC will see nothing.

We choose the Active/Passive scenario for management simplicity. In the Active/Active scenario, if an error is made in the mapping of the interfaces to the right "fabric", redundancy is broken and whats worse: you don't even notice it until it happens.