Testing for Scalability - Expert Oracle RAC Performance Diagnostics and Tuning - page 93

If industry standard test results are available, why perform another benchmark? Industry benchmarks are done
on a specific set of hardware configuration and it's not necessary that all components used in the test are used in the
cluster here. These tests are only to validate and tune the current configuration for scalability.

■ in this specific workshop example, we discuss a Datawarehouse implementation. the steps and procedures
should be similar for an oltp or a mixed workload configuration.

Note

Step 3
As covered in the capacity planning discussion of Chapter 2, it's important to understand the ratio of user workload
to the number of instances and the workload that can be handled per server. In Chapter 2, the tests focused on
determining what kind of hardware is required; Phase II tests are to make sure the components configured are right
and the parameters are sized and tuned well.
Scalability tests should start with a single node and other nodes should be added to the configuration through
iterations. The first server becomes the initial baseline; and as new nodes are added to the configuration, bottlenecks
identified and fixed, they become the new baseline for the test. Every time a baseline is determined, the new goal
is set to improve the baseline in a subsequent test until there is a point of no return, meaning no improvements in
scalability numbers.
Using the Hammerora tool (Figure 4-3 ), the initial testing goals are defined and listed in Table 4-1 .

Table 4-1. RAP Phase II—Hardware Scalability Load on One Server

Test # Scalability Factor No. of Users No. of Nodes Iterations Duration

1 1 20 1 5 1 hour

During the test, the CPU load average, the I/O load average, and the response times were acceptable. There was
no indication on any component in the configuration that may require any changes.

Step 4
With a good satisfactory load average on the one instance of the database, the test can be repeated or workload can
be increased after adding additional nodes to the cluster. So far it was a single instance configuration, and none of the
RAC components were involved. In this test, as we move to a two-node configuration, the interconnects will be used
and the storage system will be shared by more than one node. Table 4-2 illustrates tests with same scaling factor but
with more user workload.

Table 4-2. RAP Phase II—Scalability Load on Two Servers

Test # Scalability Factor No. of Users No. of Nodes Iterations Duration

2 1 30 2 5 1 hour

In Test 1,Hammerora was able to average about 12,000 queries per hour; and now with multiple servers and
increased user workload, the queries per hour almost doubled. Figure 4-4 illustrates the transaction counter from the
Hammerora tool illustrating an average workload of 22,680 queries per house.