||Oracle Tips by Burleson
Given the timing issues already present in the ATA
testing environment, the decision was made to not switch to archive
log mode and table and index level logging for the test runs. In
addition, the query stream was modified to allow reconnection of the
test user between runs to force release of the undo and temporary
segments. Why this was required in the ATA testing and not in the SSD
testing is unclear but it is probably due to the excessive time
required to run most of the ATA queries.
The ATA array was tested for the following
Nologging and noarchive base run
Nologging and noarchive
Logs and temp files on SSD
Data on SSD full buffer memory (1 gigabyte)
Data on SSD half buffer memory (500 megabytes)
The results for the queries are shown in Figure
4.2. The percentage difference between SSD and SATA is so large it is
not feasible to report them. However, the total elapsed times, adding
in a base of 1440 minutes per each non-complete query for the SCSI,
show a factor of 179 times difference between the SSD and ATA results
in favor of the SSD drives. Remember that the queries for the SSD
tests all completed, while several in the ATA tests were halted at the
24 hour point.
Please note that in order to plot the results on a
single plot the vertical scale had to be switched to logarithmic.
Therefore the changes in query time may only appear to be slight on
the graph but may actually be several tens of percentage points.
Review the detailed results in Appendix D. The runs not involving the
SSD showed consistent results.
When the temp and undo tablespaces were moved to
the SSD drives some queries improved while others got worse. When the
data tablespace was moved to SSD all query times were improved by an
average of 100 percent.
The reduction of
by half with the data tablespace
on SSD actually shows some improvements in performance on queries 2,
8, 11, 13, 13a while the others performed worse than the full memory
levels. However all of the queries performed better with the data on
the SSD than the ones on 100% ATA.
Needless to say, concerns about SSD performance
were laid firmly and completely to rest when the first query on the
SCSI and ATA tests had to be aborted at the 30 hour+ point. While at
least a 30% improvement in query speeds was expected based on the
previous insert and update based tests, it was a very pleasant
surprise when the factor of 176 times improvement for overall query
time was revealed.
Next it is time to compare the various SSD and ATA
The above book excerpt is from:
Solid State Disk Tuning
High Performance Oracle
tuning with RAM disk
Donald K. Burleson & Mike Ault