I am trying to correlate the db_block_size and
db_file_multiblock_read_count. I have the block size of 8k. Is 16k block
size equivalent to increasing the db_file_multiblock_read_count (I know
internally not but throughput wise)? I believe the I/O
size=db_block_size * db_file_multiblock_read_count. So to match the I/O
size, I could increase either one, right? Then why is it important that
Oracle store only in multiples of OS block sizes?
In the case of full table scans, your I/O size calculation is correct. However, remember that Oracle frequently reads single blocks from disk. Your I/O size in those cases would be just the size of an Oracle block.
You want the Oracle block size to be at least the size of an OS block
because you don't want to waste I/O bandwidth. For example, if the OS block
size is 8k, but your Oracle block size is 4k, you'd be transferring 8k to
get 4k into your buffer cache, in effect ignoring the other 4k that you
read. The same applies to writes.
This was first published in June 2004