Using Serviceguard Extension for RAC, 6th Edition, April 2008

As shown in Figure 2-3, each primary and standby pair protects against a single failure.
If the subnet with SG-HB (lan1/lan2) fails, Serviceguard will resolve the subnet failure
with a Serviceguard cluster reconfiguration. If the 192.168.2.0 subnet (lan3 and lan4)
fails, Oracle instance membership recovery (IMR) will resolve the interconnect failure
subnet, unless Serviceguard subnet monitoring is used. Oracle will wait for IMR time
interval prior to resolving the subnet failure. In SGeRAC configurations, default value
of IMR time interval may be as high as seventeen minutes.
CLUSTER_INTERCONNECT_SUBNET can be configured for RAC instance MNP to
monitor the RAC-IC subnet that is different from CSS-HB subnet. The parameter file
(SPFILE or PFILE) for RAC instances must have cluster_interconnects
parameter defined, to hold IP address in the appropriate subnet if the RAC Instances
use a RAC-IC network different from CSS-HB network. No special subnet monitoring
is needed for CSS-HB Subnet because Serviceguard monitors the subnet (heartbeat)
and will handle failures of the subnet.
The database instances that use 192.168.2.0 must have cluster_interconnects
defined in their SPFILE or PFILE as follows:
orcl1.cluster_interconnects=’192.168.2.1’orcl2.cluster_interconnects=’192.168.2.2’
NOTE: Do not configure CLUSTER_INTERCONNECT_SUBNET in the RAC instance
package if the RAC-IC network is the same as CSS-HB network.
The following is an example of the relevant package configuration parameters:
Oracle RAC Instance PackagePACKAGE_NAME RAC_PACKAGEPACKAGE_TYPE
MULTI_NODELOCAL_LAN_FAILOVER_ALLOWED YESNODE_FAIL_FAST_ENABLED
NOCLUSTER_INTERCONNECT_SUBNET 192.168.2.0
NOTE: For information on guidelines to change certain Oracle Clusterware and
Serviceguard cluster configuration parameters, see “Guidelines for Changing Cluster
Parameters” (page 53)
Faster Failover Configuration (SGeFF and SGeRAC)
Faster Failover configurations with SGeFF require at least two separate SG-HB networks,
and are restricted to two nodes.
Cluster Communication Network Monitoring 51