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

• It allows one NIC to be a standby for multiple NICs which can reduce the total IO slots required

relative to the APA.

• It supports the failover to different NIC types in the same family, like failover from 100B-T Ethernet

to 1 Gb Ethernet.

• It monitors all links and can detect the loss of all links and take failover action.

HP-UX APA supports two configuration modes: Failover Group and Link Aggregate.

APA Failover Group mode is supported in RAC with SGeRAC but has the following

disadvantages:

• It only monitors the health of the standby NIC in the same system.

• It requires one standby NIC per one active NIC.

• It only supports the failover of the same NIC type.

• It doesn’t monitor the loss of all links; they must be detected (after a timeout period) by the

application.

• In some failure cases, APA failover from an active to standby LAN can take up to 30 seconds.

Beginning with HP-UX 11i v3 and 11i v2 September 2007 or later, the APA HA failover mode,

LAN_MONITOR is supported with Serviceguard. This is supported beginning with Serviceguard

A.11.17.01 on HP-UX 11i v3, and Serviceguard A.11.17 with the patch, PHSS_35427 or later on

11i v2.

APA Link Aggregate mode provides the ability to combine the speed of multiple NICs into one

single pipe (called a trunk) for larger bandwidth.

Cluster Interconnect Subnet monitoring

This mode requires all APA links connected to the

same network switch. This mode is not recommended due to the network switch being a SPOF in this

configuration. If the APA Link Aggregate mode is required for higher bandwidth, it should be

configured with multiple trunks in a primary/standby configuration in conjunction with Serviceguard

primary/standby LANs feature for high availability. Only 100Base-T and Gigabit cards are supported

in an APA aggregate on HP-UX 11i v1, 11i v2 and 11i v3. 10-Gigabit cards are supported with APA

LAN_MONITOR mode only.

Starting with A.11.18, SGeRAC introduced a new feature called Cluster Interconnect Subnet (CIS)

monitoring to improve failure recovery time (over a pure Oracle RAC environment) and to address a

limitation with subnet monitoring, which may cause all nodes to fail in the event of a double switch

failure. The feature adds a new parameter, CLUSTER_INTERCONNECT_SUBNET, for use with

multi-node packages (MNPs). When RAC instances are configured with MNP packages and the RAC

subnets are monitored using CLUSTER_INTERCONNECT_SUBNET, if the monitored subnet fails, at

least one MNP package (i.e., the RAC instance) will continue to run.

The use of CLUSTER_INTERCONNECT_SUBNET is strongly recommended, to provide continued

functionality in the event of a double-switch failure.

For information, please see

Using Serviceguard Extension for RAC, 9

edition.

The load balancing capability of APA is highly dependent upon the type of traffic over the aggregate and the configured load balancing

algorithm. For more information on APA load balancing, see the

HP Auto Port Aggregation Administrator’s Guide.