User guide

146 Implementing Cisco InfiniBand on IBM BladeCenter
7.6 Implementing a design using gateway redundancy
In many cases, customers want to make sure the InfiniBand hosts have High Availability (HA)
connectivity to the outside world. One element of this HA environment is redundant Ethernet
gateway modules to the Ethernet network.
7.6.1 Some considerations for redundancy
When implementing a design using gateway redundancy, remember the following
considerations:
򐂰 In this example, both gateway modules are in the same 3012. For maximum HA, it would
be desirable to have each of the gateway modules in the redundancy group in two
separate 3012s. Cross-3012 Ethernet Gateway redundancy is supported, but we are not
going to demonstrate it in this redundancy example.
򐂰 If placing both gateway modules in the same 3012, it is best to have one gateway module
use its internal port to the top InfiniBand module (internal port 1) and the other gateway
module use the port to the bottom InfiniBand module (internal port 2). This will permit
maximum redundancy
򐂰 The InfiniBand management interface (int mgmt-ib) must be configured and up (the no
shutdown command must be present in the config for the mgmt-ib interface) for
redundancy to work. This interface is used by the redundant gateways to keep track of
each others operational state
򐂰 When using gateway load-balancing (active-active mode), each bridge group must have
an IP assigned to it. This is not necessary if using active-standby or non redundant
operation
Tip: At a minimum, you should use the InfiniBand port 2 for connection to the
InfiniBand fabric. This is because the design of the 3012 is such that the module in
slot 15 (InfiniBand port 1 on the Ethernet gateway) will not work if the InfiniBand
module in slot 16 is not present or otherwise in a failed state. To state this a different
way, an InfiniBand module in slot 16 will work by itself, but an InfiniBand module in
slot 15 will only work if there is an InfiniBand module in slot 16.