HP Serviceguard Extension for RAC (SGeRAC) manual SG-HB/RAC-IC Traffic Separation

Alternate Configuration—Fast Reconfiguration with Low Node Member Timeout

High RAC-IC traffic may interfere with SG-HB traffic and cause unnecessary member timeout if Serviceguard cluster configuration parameter MEMBER_TIMEOUT is low. If MEMBER_TIMEOUT cannot be increased, use of an additional network dedicated for SG-HB alone avoids unnecessary member timeouts when RAC-IC traffic is high. This configuration is for a cluster with two or more nodes that have high RAC-IC traffic and/or need faster failover (a low value for Serviceguard configuration parameter MEMBER_TIMEOUT).

NOTE: Starting with Serviceguard A.11.19, the faster failover capability is in core Serviceguard. This configuration can be used for faster failover.

Figure 8 SG-HB/RAC-IC Traffic Separation

Each primary and standby pair protects against a single failure. With the SG-HB on more than one subnet, a single subnet failure will not trigger a Serviceguard reconfiguration. If the subnet with CSS-HB fails, unless subnet monitoring is used, CSS will resolve the interconnect subnet failure with a CSS cluster reconfiguration. It will wait for the CSS misscount time interval before handling the CSS-HB subnet failure (by bringing down the node on which the CSS-HB subnet has failed).

The default value of CSS misscount in SGeRAC configurations is 600 seconds.

As shown in Figure 8, CLUSTER_INTERCONNECT_SUBNET can be used in conjunction with the NODE_FAIL_FAST_ENABLED package configuration parameter to monitor the CSS-HB network. A failure of CSS-HB subnet on a node should be handled by bringing down that node. Therefore, set NODE_FAIL_FAST_ENABLED to YES for the package monitoring the CSS-HB subnet. If the monitored subnet fails, the failure of the CSS-HB subnet on a node will bring down the instance of the multi-node package and the node where the subnet has failed (When Oracle Clusterware is configured as a multi-node package and CLUSTER_INTERCONNECT_SUBNET is used to monitor the CSS-HB subnet).

A failure of CSS-HB subnet on all nodes will result in the multi-node package failing on the nodes one by one (resulting in that node going down), and one instance of the multi-node package and node will remain providing services to the clients.

Use a separate package to monitor only the CSS-HB subnet and have Oracle Clusterware multi-node package depend on the package monitoring the CSS-HB subnet. The NODE_FAIL_FAST_ENABLED parameter is set to NO for the Oracle Clusterware package, and is set to YES for the package

Cluster Communication Network Monitoring 37