Re: VxFS Block Size

Hello Steve,

I'll try to change my question. Is there sense to make db_block_size less than memory_page_size, e.g. db_block_size=2K and memory_page_size=4K under NT or db_block_size=8K and memory_page_size=16K under Relaint UNIX (as I know these memory page sizes are constant). Of course, such memory page size does not constrain the size of I/O operations but I concern about possible 50 % memory wastage because of db_block_size=1/2 of memory_page_size in these cases. Or Oracle works with memory itself without taking OS memory page size into consideration ?

Regards,
Slava.

In article <39fd68b3.382571959_at_nsw.nnrp.telstra.net>,   steve.adams_at_ixora.com.au (Steve Adams) wrote:

> Hello Slava,
>
> You need to know which path through the I/O subsystem is being used,
 and which
> components of that path constrain the size of I/O operations.
>
> The most important is the file system. UFS style file systems
 constrain I/O to
> the file system block size. Operating system buffered I/O to an
 extent based
> file system like VxFS is constrained to whatever the standard file
 system buffer
> size is. I don't know Reliant Unix, so I cannot tell you what the
 number is for
> you, but you should be able to work it out for your self by studying
 the ratio
> of 'blks/s' to 'r+w/s' in the 'sar -d' output for various
 filesystems. If you
> use direct I/O or raw, then the file system buffer size is not
 relevant.
>
> The standard file system buffer size is typically the same as the
 memory page
> size or page cluster size, however, there is no necessary
 relationship - it is
> just a natural choice. Thus the page size should not be regarded as a
> constraint. Incidentally, could you please confirm that memory page
 size with
> getconf PAGESIZE
>
> If you are using a logical volume manager product, then you need not
 worry about
> its concept of the page size, however, I/O operations that span a
 logical track
> group boundary will be split at that boundary.
>
> If you have disk striping, then obviously I/O operations will be
 split at stripe
> element boundaries. Thus the alignment of database blocks to stripe
 elements is
> important.
>
> I am not aware of any cached disk arrays that constrain physical I/O
 operations
> to their cache slot sizes.
>
> @ Regards,
> @ Steve Adams
> @ http://www.ixora.com.au/
> @ http://www.christianity.net.au/

Sent via Deja.com http://www.deja.com/
Before you buy. Received on Mon Oct 30 2000 - 09:16:05 CST