Optimize Distributed Workload - Expert Oracle RAC Performance Diagnostics and Tuning - page 502

When the PMON updates the listener with the profile statistics, it also makes an entry in the listener log file. This
could be tracked by the service_update string. The frequency of update could also be tracked using the timestamp
found against the service_update entries. For example, the following output indicates that the PMON has been
updating the listener:

02-FEB-2014 23:06:50 * service_register * +APX1 * 0
03-FEB-2014 18:18:56 * service_register * LsnrAgt * 0
03-FEB-2014 18:19:01 * service_register * -MGMTDB * 0
03-FEB-2014 18:19:15 * service_register * +ASM1 * 0
03-FEB-2014 18:19:26 * service_register * -MGMTDB * 0
03-FEB-2014 18:19:43 * service_register * +ASM1 * 0
03-FEB-2014 18:22:13 * service_register * SCDB_2 * 0
03-FEB-2014 18:23:36 * service_register * +APX1 * 0
03-FEB-2014 18:23:48 * service_register * SCDB_2 * 0
04-FEB-2014 00:40:21 * service_register * SCDB_2 * 0

The load statistics available on the listener on the respective nodes is used to reroute any connection to the node
that has the least load.
The following steps are performed to reroute connection requests based on user workload:

1. A user connection is established to a listener using the client load balancing options

discussed earlier.

2. The listener where the connection was originally established will, based on the load

statistics available, reroute the connection to another listener on another node. (The
listener information is obtained from the REMOTE_LISTENER parameter).

With the introduction of services and distribution of services across various instances and based on user(s)
business requirements, load balancing criteria will vary. This will depend on having a symmetric distribution or
an asymmetric distribution of services and on the capacity of the nodes participating in the cluster. For symmetric
services and nodes with similar capacity, the absolute session count by instance evenly distributes the sessions across
the nodes; and if the service distribution is asymmetric or the nodes do not have a similar capacity, then the run
queue length of the respective nodes is used to determine the least loaded node.
Oracle provides the DBAs the option of defining goals and determining load-balancing criteria. Load balancing
goals could be:

1. Based on elapsed time: Under this method, a new ranking referred to as goodness of

service is used in the load-balancing algorithm. Under this method, the load balancing
is driven by the actual service time that would be experienced by the session on a given
instance. Ranking compares service time, referred to within the database as the metric
“ Elapsed Time Per User Call .”

2. Based on the number of sessions: Under this method, the load across the various nodes is

balanced based on the number of Oracle sessions connected to the database. In this case,
the actual resource load or response time or service time is not considered. However, the
basic count on the number of sessions is considered to determine the least loaded node
and where the next session should be connected.

$GRID_HOME/bin/srvctl modify service -d SCDB -s FIPS -clbgoal SHORT -rlbgoal
THROUGHPUT -notification TRUE