Serviceguard Network Manager: Inbound Failure Detection, March 2007

in each node, the two NICs remotely poll each other. If the NIC in Node B fails and is identified as
failed, and there is no traffic other than the remote polling traffic that just stopped, the inbound
statistics of lan0 on node A will not be incrementing. Using the INONLY_OR_INOUT setting,
Serviceguard Network Manager running on node A will declare lan0 on Node A down— the whole
subnet will be down. In this situation the default setting of INOUT would be a better choice. See
Figure 6 for an illustration of this example.
Figure 6. No standby NICs, single switch, NIC fails
Guidelines and recommendation
As long as the network configuration is highly available, users should be able to avoid the situations
described in the previous examples. The following guidelines are strongly suggested when evaluating
use of the INONLY_OR_INOUT setting for network failure detection.
Test your cluster thoroughly before and after setting the parameter to INONLY_OR_INOUT. Run the
application on the cluster to be sure the cluster network configuration is suitable for the option. The
network environment where the INONLY_ORINOUT setting is applied has considerable impact on
Serviceguard Network Manager. Be sure that the following conditions are met before changing from
the default INOUT setting:
• All local bridged networks in each cluster member node have at least two interfaces.
• Each primary interface has at least one standby interface connected to another.
• The primary switch is directly connected to its standby.
• The standby switch is connected to a router through a different path other than the primary switch.
• There is no single-point-of-failure anywhere in all of the bridged networks
When the INONLY_OR_INOUT setting for failure detection is applied in an appropriate network
environment, users will be able to avoid hanging applications caused by inbound-only failures when
NICs or switches fail without risks.
8