Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: VxFS Block Size

Re: VxFS Block Size

From: Steve Adams <steve.adams_at_ixora.com.au>
Date: Mon, 30 Oct 2000 00:07:27 GMT
Message-ID: <39fcb39b.336211336@nsw.nnrp.telstra.net>

Hi Greg,

The VxFS block size has to do with space allocation. Specifically, there is one bit in the extent bitmap area for each data block in the filesystem. Extents can be allocated to files in extents of any binary power between 1 and 2048 blocks. Unless direct I/O is being used, the size of I/O operations is the standard file system buffer size (typically 8K) and has nothing to do with the VxFS block size.

The VxFS defaults are intended for general purpose filesystems. For VxFS filesystems dedicated to Oracle datafiles, if the small default VxFS block size is used, then the extent bitmap area will be larger than necessary and the number of extents needed per datafile will be unnecessarily large too. This has a noticeable effect on datafile creation performance, and a small impact on Oracle multiblock reads too.

Incidentally, the fragmentation of large files increases when file system free space falls below 10%. So datafiles that may be subject to multiblock reads are best created first in their filesystem, rather than last.

@ Regards,
@ Steve Adams
@ http://www.ixora.com.au/
@ http://www.christianity.net.au/

-----Original Message-----
From: Greg Stark <greg-spare-1_at_mit.edu>

Oddly enough Veritas recommends 2k block size.

Conceivably the smaller the block size the more effective the Oracle buffer cache and veritas cache can be. But it still feels weird. Received on Sun Oct 29 2000 - 18:07:27 CST

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US