Optimize Distributed Workload - Expert Oracle RAC Performance Diagnostics and Tuning - page 503

Applications Using Connection Pooling
For applications using connection pooling, Oracle provides a more robust, cleaner, and proactive method of load
balancing called run-time connection load balancing (RCLB).
Instead of a using the old methods, where the application makes a connection to the database server to
determine the actual load on the system, under the new method, events are used to notify the application regarding
the load. Based on this information, connections are established to the least loaded machine.
RCLB relies on the ONS event mechanism (discussed in the previous sections) and FCF in applications using
Java, and OCI or ODP.NET subscribe via Oracle's advanced queuing feature. The RCLB feature provides assignment
of connections based on feedback from the instances in the RAC cluster. The connection cache assigns connections
to clients based on a relative number indicating what percentage of requested connections each instance's service
should handle. It is enabled automatically when RAC starts to post service metrics. Service metrics provide service
levels and percentage distributions for each instance of a service. Connections to specific instances in the cluster are
based on the service metrics available.
Oracle uses the services metrics values calculated and stored in the automatic workload repository (AWR) to
determine current load characteristics on the various instances. The service metrics is thus forwarded to the master
MMON background process. The MMON in turn builds the required load advisory and posts the required advice to AQ,
PMON , and the ONS.
Notification mechanisms are based on one of two definitions:

1. Service time measures the elapsed time vs. the demand. When this option is selected,

Oracle examines all of the time consumed in the service from an efficiency and delay
perspective and rates this data against the service-level goals set for the service. Using
service time or response time for load balancing recognizes machine power differences,
sessions that are blocked in wait, failures that block processing, as well as competing
services of different importance. Using the proactive propagation method ensures that
work is not sent to overwork, hung, or failed nodes.

2. Throughput measures the efficiency of the system rather than delay. Throughput

measures percentage of the goal response time that the CPU consumes for the service.
Basically, throughput is the number of user calls completed in a unit of time.

■ rClb of work requests is enabled by default when FCF is enabled. there is no additional setup or configuration
of ONs required to benefit from rClb.

Note

Load Balancing Definition
Load balancing can be configured using the OEM or srvctl command line utility. Connection load balancing is
enabled by setting the CLB_GOAL parameter to appropriate values.
Valid values for CLB_GOAL are listed in Table 15-4 .

Table 15-4. Client Load Balancing Parameters

Goal Type Value Description

CLB_GOAL_SHORT

1 Connection load balancing based on elapsed time.

CLB_GOAL_LONG 2 Connection load balancing based on number of sessions.