Re: Schema Placement for Purchased Apps

"Gabriel Gonzalez" <no-spam_at_no-spam.com> wrote in message news:1u6cnXbELPnB4AijXTWcrg_at_giganews.com...
> > They're kinda equivalent when performing multi-block reads but what
about
> > (the typically more common) single block read ? And perhaps more
>
> When would you have a single-block read? My understanding (again, if I'm
> wrong let me know!) is that Oracle only reads "multi-block" blocks at a
> time, never less. It puts the excess blocks into the cache, should they
be
> needed.

No: index leaf nodes, for example, are always read one block at a time.

>
> > importantly, what about the writes DBWRs needs to perform back to the
O/S
> > system ?
>
> Does it matter? If you need to commit to disk an Oracle 2k block vs a 64k
> block, at the physical level it is still a whole-physical-block operation
> (when you get down to physical writes, the O/S can only handle a block at
a
> time, even if you are only changing one bit out of the whole block).
>
> > rotation latency) the general recommendation is to match your block size
> > with the I/O buffer size.
>
> So since the physical block size of most systems is 64k, we should be
using
> 64k Oracle blocks everywhere?

Er, the physical block size of most systems is 512 bytes.

*Please* read Steve Adams' material on this issue! He'll even tell you that AIX and Linux should use 4K blocks, HP-UX and Solaris should use 8K blocks.

Regards
HJR
>
>
Received on Thu Apr 10 2003 - 16:54:53 CDT