Re: Oracle block size - OS block size

Posting such information without OS and its version information is almost useless. For instance, I can eassily prove that to be wrong if thought to be universally true:

1. Cockcroft, R. Pettit, "Sun Performance and Tuning", 2nd, PTR/PH, p. 309: "a disk block is 8 Kbytes, an inode is about 300 bytes."

In fact you can find such claims in that book in multiple places.

Secondly, associating nbpi with db_block_size doesn't seem to make sense. Inodes are only fetched and searched at the start of open(2). The big portion of time reading from or writing to datafiles is read(2) and write(2) (or their variants such as pread64). Those operations do not depend on location of the inode in question. If you want to tune the time to find the inode, make sure your datafile name is less than 30 characters long so they can be cached in DNLC (directory name lookup cache (14 characters for SunOS4), tune inode cache... But as I said, it's probably a waste of time focusing your attention on this issue.

Yong Huang
yong321_at_yahoo.com

"koert54" <k_at_k.com> wrote in message news:<LyuW6.36383$mR5.4796351_at_afrodite.telenet-ops.be>...
> 8192 = nbpi or number of bytes per inode
> always make your nbpi equal to db block size - less translation overhead
>
> cheers
> Kurt Van Meerbeeck
Received on Sat Jun 16 2001 - 18:02:05 CDT