Architecture considerations and best practices for architecting an Oracle RAC solution with Serviceguard and SGeRAC

RAC:

• RAC interconnect network

If all RAC interconnect networks fail for a node, RAC evicts from the group the instance running on

the impacted node. The time to eviction of the instance is dependent on the IMR value; the default

value is 17 minutes. In this case, only the RAC instance is down and not the node; therefore, the

cluster is not impacted.

As you can see, if Serviceguard detects the failure of a resource that it monitors, Serviceguard will

provide the fastest cluster recovery, and the new cluster membership is propagated to Oracle

Clusterware and RAC. The design principle is to configure Oracle Clusterware and RAC critical

resources to be the same as Serviceguard’s critical resources and be monitored by Serviceguard. For

example, configure RAC interconnect and CSS heartbeat on the same network as Serviceguard

heartbeat, and configure the voting disk to be monitored by the Volume Monitor, or as an EMS

resource that is monitored by the HA Disk Monitor, depending on the Serviceguard version

Network HA configuration choices

used.

In a RAC configuration there are three networks and a minimum of three IP addresses configured per

cluster node:

• Private network for Oracle Clusterware CSS heartbeat

One IP address is required if configured on its own subnet. Oracle only supports one active

sub-network for the CSS heartbeat. Typically it would be configured on the same network as the

RAC interconnect. There is more discussion on this in a later section.

• Private network for RAC interconnect

One IP address is required if configured on its own subnet. Multiple active sub-networks can be

configured for RAC interconnect, but they are only used for load balancing. The primary interface

must be functioning for RAC to continue on the node.

• Public network for client connections to the RAC database

This network requires two IP addresses: one station IP address is for the purpose of system

administration, and one VIP address is for a client connection to the RAC instance running in the

system. The VIP address is required per node and is managed and monitored by the Oracle

Clusterware CRS component.

The Oracle VIP and Serviceguard package RIP can be configured on the same NIC.

The HA Disk Monitor supports S/LVM resource, but is not supported on 11i v3 systems running Intel

Itanium

9300 Series processors. Starting

SG/SGeRAC A.11.18, use the Disk Monitor for VxVM/CVM Volumes. Starting SG/SGeRAC A.11.20 with patch PHSS_41225 or later, the

Volume Monitor can be used as an alternative to monitor VxVM/CVM Volumes and LVM logical volumes (LVs).

Required Oracle 10gR2 Patch 2 (10.2.0.2)