The warehouse I’ve been working on has been struggling with IO performance for a while so I decided to poke my nose into what was going on with the setup of the database.

The warehouse uses a 32K block size and db_file_multiblock_read_count of 32 which should match the IO subsystem stripe size (?). I tried the trick of full table scanning a large table whilst running an extended trace (level 8) and it showed that indeed 32 blocks were being read in each IO request – thanks to Connor Mcdonald for the method for this – I’m sure I’ve seen it in other places too but that’s where I got it from.

A colleague of mine then passed me a URL for a page on Steve Adams site – http://www.ixora.com.au/tips/buffered_block_size.htm reading through this and the linked pages resulted in me checking various ED isn’t viewed as a characteristic piece of maturing. canada from generic viagra When you take this anti ED drug, you inadvertently take heart tonic. viagra professional australia check over here With male enhancement, you give your body a constant flow of capable sexual supplements. on cialis line Spinal decompression is successful in treating back pain, bulging or herniated check out for more info tadalafil uk buy discs, degenerative disc disease, posterior facet syndrome or radiculopathy (damage to spinal nerve roots). things.

OS is HP-UX which can support Direct IO – http://www.doag.org/pub/docs/sig/sap/2004-03/Dorer1.pdf

I Issued the mount command which showed that the filesystems were indeed mounted with convosync=direct and qio (Veritas Quick IO)

I checked the value of filesystemio_options in the Oracle database and it was set to asynch – Eureka I thought…that must be it….but after some simple back to back tests it had not improved the IO of a sample full table scan query at all.

Still a mystery…more time on it next week.