Free Oracle Tips   HTML Text

BC Oracle tuning

Oracle training

Oracle support

Remote Oracle

Oracle Tips by Burleson

SSD Benchmark Hypothesis

Now that the issues surrounding SSD in an Oracle environment have been covered, it is time to hypothesize about the results of the benchmark tests. 

In reviewing the existing research and applying logic to the results, large performance gains can be realized by placing specific types of databases on SSD. Specifically databases that are predominantly read based, such as decision support (DSS) or data warehouse, should benefit the most. In order to determine if this is indeed the case, use the TPCH benchmark, which is designed for DSS testing.

The central issue of this study is to determine the benefit of SSD for Oracle under these conditions:

No SSD with a super-large Oracle data buffer – This can minimize PIO with only one physical read into the data buffer, but cause LIO overhead.

SSD with a tiny Oracle data buffer – This makes PIO faster, but will cause repeated LIO overhead as data blocks are aged-out of the Oracle data buffer.  It is expected that LIO overhead would be huge with high latch contention, buffer busy waits, and free buffer waits.

SSD for files with large Oracle Data buffer cache – This is duplicating RAM but it would relieve the LIO issue.

Assuming that SSD is far faster for PIO, the issue of duplicitous Oracle activity still remains.  At PIO time, the data block must still be read into db_cache_size and then a consistent get is required to deliver the data block to the requesting program. Hence, RAM resources appear to be better allocated directly to db_cache_size instead of using the RAM with SSD.

Specific architectures

Large data buffers with SSD for:

• Redo log files

• Undo segments

• Temporary tablespace blocks

Predictions

If it is assumed that SSD is far faster for PIO, the issue of duplicitous Oracle activity still remains.  At PIO time, the data block must still be read into db_cache_sizeand then a consistent get is required to deliver the data block to the requesting program. Hence, RAM resources appear to be better allocated directly to db_cache_size instead of using RAM with SSD.

The benchmark study that was created specifically to test these hypotheses and show conclusively how Oracle functions with solid-state disk will follow in the next Chapter.

The above book excerpt is from:

Oracle Solid State Disk Tuning

High Performance Oracle tuning with RAM disk

ISBN 0-9744486-5-6  

Donald K. Burleson & Mike Ault

http://www.rampant-books.com/book_2005_1_ssd.htm