Sample Configurations with SGeRAC and Oracle RAC 10gR2, March 2009

has sufficient bandwidth. If the primary network fails, Serviceguard performs a local LAN failover to

use the standby network. Node failure is detected when Serviceguard misses heartbeats.

Configurations with heavy RAC-DB-IC traffic may place a limit on how aggressive the Serviceguard

heartbeat timeout can be used since SG-HB may not be processed in time. Therefore, a longer

Serviceguard heartbeat timeout may be needed to avoid false cluster reconfigurations.

Figure 2. Dual SG-HB with CSS-HB and RAC-DB-IC on same subnet

Figure 2 shows one method to overcome the heavy RAC-DB-IC traffic affecting SG-HB. This

configuration uses two SG-HB networks. Two SG HB networks are required in faster failover

configurations. The Serviceguard heartbeat timeout can be as aggressive as the software

configurations allow. The subnet for CSS-HB and RAC-DB-IC has redundancy with primary and

standby adapters and switches configured. If the primary (lan1) fails, Serviceguard performs a local

LAN failover to the standby (lan2). If a node fails, Serviceguard detects the failure from the loss of

SG-HB.

If both primary (LAN 1) and standby (LAN 2) fail, Serviceguard logs the failure but will not take

action unless Serviceguard packages with monitored subnets are configured to take action, for

example node failfast. The advantage for Serviceguard to take action on a monitored subnet is that

the failure detection can be faster than the CSS-HB timeout, and thus recovery action can be quicker.

Use of Serviceguard subnet monitoring has a limitation where if all interconnect fails (for example

primary and standby switch failed at the same time), all the nodes are halted. If there is a concern

with simultaneous failure of both switches, starting with SGeRAC A.11.18, Serviceguard supports

cluster interconnect subnet monitoring.

If there are no Serviceguard packages configured to take action, CSS-HB timeout occurs and CSS

goes through reconfiguration and reboot the node that is not part of the CSS cluster membership.

When the node reboots, Serviceguard will reform with the new membership.

Node A

LAN 1

Node B

LAN 3

CSS-HB

RAC

Private

(primary)

Private

(standby)

Private

(primary)

LAN 1

LAN 2