Sizing Exadata - Oracle Exadata Recipes: A Problem-Solution Approach - page 109

The redo writes statistic is slightly different. This statistic is incremented each time LGWR successfully flushes
the contents of the log buffer to disk. Contrary to the physical read IO requests and physical write IO requests
statistics, redo writes is a reflection of “completed I/O,” not I/O demand. However, for purposes of conducting a IOPs
analysis, it is often acceptable to treat this statistic the same as the read and write request statistics.
Physical write and redo write IOPs numbers as measured from Oracle system statistics ( V$SYSSTAT or
DBA_HIST_SYSSTAT ) are reflections of Oracle's write calls but do not account for the fact that extents will be mirrored
with either ASM normal or high redundancy. When sizing for Exadata, you need to multiply the system statistics by
either two for normal redundancy or three for high redundancy disk group configurations.
Since many of statistics are counter-based metrics that continuously increment over time, you need to subtract
the previous snapshot's value from the current snapshot's value and divide by the number of seconds between
snapshots to calculate your I/O bandwidth rates. In this recipe, we used Oracle's LAG function to achieve this. We also
used 11g PIVOT method to dynamically build a list of aliased columns for each of the metrics we wanted to capture.
We like this approach because it reduces the number of lines of code and reduces self-joining potentially large
AWR tables.

■ If your current databases are experiencing significant redo log write wait activity, the redo write statistic
may not be an entirely accurate picture of redo writing IOps demand. If you experience significant wait time for
log file sync waits it can be an indication of an I/O issue or configuration issue on your current environment.
Exadata can help address this with its disk cache on the storage servers and Smart Flash Logging. to learn more
about these features, please see recipe 18-9.

Caution

While the methods presented in this recipe should help you to properly size your environment for Exadata, other
factors also play a significant role. CPU processor capacity planning, I/O bandwidth capacity planning, and certainly
storage capacity planning are other important tasks. Please read the other recipes in this chapter to learn more about
these capacity-planning activities.

5-3. Determining I/O Bandwidth Requirements
Problem
You wish to determine whether an Exadata Database Machine will provide enough I/O bandwidth, or data
transfer capacity, to meet your database workload requirements. Furthermore, you wish to decide which Exadata
configuration will meet your needs based on information gleaned from AWR data.

Solution
To conduct an Exadata I/O bandwidth capacity-planning exercise, we recommend the following approach:

• Query raw I/O bandwidth information from Oracle AWR, export to Excel, and optionally plot

trends and/or identify outlier conditions.

• Determine I/O bandwidth averages, maximums, and other statistical information to

understand usage patterns.

• Add I/O bandwidth numbers for additional Oracle databases targeted to be deployed on

Exadata.

• Determine Exadata and Exadata model specific fit based on measured I/O bandwidth

from AWR.