Workload Qualification - Oracle Exadata Recipes: A Problem-Solution Approach - page 76

At this point, assuming the workload depicted in Figure 4-2 is representative of a typical load, you have some
data points to help facilitate your Exadata fit assessment. But to make intelligent decisions about the data, you need to
understand what business processes or functions these SQL statements are associated to, and more so, whether these
are important to optimize. This is not a technical analysis. Rather, it's a business analysis supported by metrics and data.
Next, take a look at the Segments by Physical Reads section of your AWR report, as displayed in Figure 4-3 .

Figure 4-3. Segments by Physical Reads

The information in Figure 4-3 requires an even deeper understanding of how physical I/O on these objects is
related to overall performance for key business objectives.
The points we're attempting to illustrate are as follows:

• Use data from your existing database's workload repository to understand what your I/O

workload profile is.

• Determine whether you have important business functions whose throughput or response

time is limited by physical I/O.

• Furthermore, determine whether the queries accessing this data are performing full scans to

retrieve the data.

• Do not rely on macroscopic, system-level statistics to drive your decision. Start by looking at

the SQL statement level and ensure that you've got a collection of “important” SQL statements
that is frequently performing full scans, is I/O-bound, and requires performance improvement
to enable business success or reduce business risk.

What about Exadata's other software features? Table 4-1 provides brief explanations of how we recommend
conducting an Exadata fit assessment based on your current database.

Table 4-1. Exadata Storage Software workload fit assessment

Software Feature What to Look for in Current Database Exadata Fit Assessment Comments

Hybrid Columnar
Compression

Very large tables with static data that
undergo near-zero DML

HCC is typically not an Exadata “driver,” but in
certain situations it can be depending on data
volumes and DML patterns. If disk capacity is
your #1 business problem, there are probably
cheaper ways to solve it than with Exadata.

( continued )