Testing for Scalability - Expert Oracle RAC Performance Diagnostics and Tuning - page 86

3. While scaling up on a single server can create a bottleneck causing a single point of failure.

The scale-up option can also cause loss of processing power or limitations to scalability of
the application when one or more servers in the configuration fail. For instance, if one out
of the four servers in a four-node RAC cluster fails, availability is reduced by 25%. This is a
substantial loss for most applications configured for prime time. Now consider that if two
or more servers fail, the availability will reduce by 50% and 75%, respectively. Based on the
workload and the active concurrent sessions, the cluster may not be useable.

Vertical

Horizontal

Figure 4-1. Scale-in and scale-out

Scale-Out
Whereas adding additional servers can increase the current resource capacity of the servers, each server within
has the capacity to accommodate additional CPU and memory if required. Scale-out configurations are normally
clustered configurations where servers are interconnected through a private dedicated network for cluster
management. Figure 4-1 illustrates a scale-out configuration. Scale-out also has certain limitations:

1. Not all user applications can run on a clustered configuration. If the application and

persistence layer (which contains all the SQL operations to the database) are not tuned
and optimized efficiently, scale-out configurations may not bring the performance
benefits expected. In many cases poorly running SQL statements could create negative
impact when executed in a clustered configuration. A certain amount of testing is required
to ensure that applications perform and scale per expectations.

2. Cost of implementing such a solution maybe much higher. The infrastructures, technical

know-how, and the learning curve required implementing this solution maybe high.

Clustering provides great benefits of distributing workloads among the various servers participating in the cluster.
RAC is a clustered database solution that is designed to help distribute workload among several servers. Added with some
of the other features available with Oracle Database 11g Release 2 and higher, RAC provides scale-up and scale-out.