Re: Database Block Size

Dave A wrote:
>
> The databases we have on QuickIO that we have engineered tests for show on
> average a 30% increase over raw. Our test mainly consists of batch inserts,
> deletes and select statements using different access paths(full table scan,
> index range scan etc.) It is true that not every transaction shows
> improvement over raw, but I have recovered databases on raw disk and that's
> enough to make me really like not dealing with raw anymore. It's also nice
> to be able to resize datafiles instead of having to add another datafile to
> grow a tablespace.
>
> My overall opinion is that if you don't want the administration chores of
> dealing with raw, you can put a db on filesystem and get as good, if not
> better performace if you use QIO.
>
> --
> Dave A
>
> "Steve Adams" <steve.adams_at_ixora.com.au> wrote in message
> news:39ee2564.2728313_at_nsw.nnrp.telstra.net...
> > Hi Frank,
> >
> > On the issue of raw, have a look at the whitepapers on the Veritas web
 site.
> > They consistently claim to get "near raw disk performance", except that
 they can
> > claim to exceed raw disk performance when using Cached Quick I/O on large
 memory
> > systems running 32-bit Oracle. In other words, unless you are totally
> > incompetent in sizing your memory allocations, raw will outperform the
 expensive
> > Veritas solutions marginally, and may outperform their standard solutions
> > dramatically if your I/O workload is heavy. See their white paper
 "VERITAS
> > Database Edition 2.1 for Oracle - Performance Brief - OLTP Comparison on
 Solaris
> > 7" at http://eval.veritas.com/webfiles/docs/dbed_2-1_perf.pdf for more
 details.
> >
> > @ Regards,
> > @ Steve Adams
> > @ http://www.ixora.com.au/
> > @ http://www.christianity.net.au/
> >
> > -----Original Message-----
> > From: "Frank" <frankbo_at_interaccess.nl>
> >
> >
> > I'm convinced. And willing to learn. And -rereading the thread in Unix
 mode-
> > it makes more sense. Or the opposite <grin>
> > Will go for 8k on Unix as from now on. Cooked, that is.
> > BTW, I come across many (well...) sysops that are _very_ reluctant to
> > start off with raw fs. Cannot convince them always.
> > Any war stories on terrible performing dbms's on cooked fs, that flash on
> > raw?
> > --
> > Frank
> > Howard J. Rogers <howardjr_at_www.com> schreef in berichtnieuws
> > 39ec3a53_at_news.iprimus.com.au...
> > > "frank" <fbortel_at_home.nl> wrote in message
 news:39EB3CD5.F8F83116_at_home.nl...
> > > > Comment inline
> > > >
> > > > "Howard J. Rogers" wrote:
> > > >
> > > > > Comments below
> > > > > HJR
> > > > > --
> > > >
> > >
> >
> > --------------------------------------------------------------------------
> > > > > Opinions expressed are my own, and not those of Oracle Corporation
> > > > > Oracle DBA Resources:

 http://www.geocities.com/howardjr2000
> > > >
> > >
> >
> > --------------------------------------------------------------------------
> > > >
> > > > > <snip>
> > > > > That's got to be one of the most sensible postings of the subject
 I've
 seen
> > > > > for weeks. All of it makes sense with (forgive me) the one
 exception
 of
> > > > > rather avoiding the discussion about block size. I keep seeing
 total
 cr*p
> > > > > posted here about the "fact" that block size would seem to depend on
 what
> > > > > you use the database for: OLTP allegedly requires a small block
 size,
 and
> > > > > data warehousing doesn't.
> > > >
> > > > Maybe you could tell Oracle to alter the documentation, then? You're
 closer
> > > > than me ;-)
> > > >
> > >
> > > Have you read the documentation for the Backup and Recovery course???
 The
> > > DBA course is bad enough, but the BUR one is worse. And no, complaining
> > > about it doesn't make a bit of difference. So whatever the
 documentation
> > > says, talk to the experts, and Steve Adams knows what he's talking about
 it
> > > (as far as I'm concerned, anyway).
> > >
> > > And he says 8K is a given for most Unixes with file systems, for reasons
> > > that make entire good sense as far as I can work out.
> > >
> > > So take it up with him.
> > >
> > > Personally, I reckon this subject arouses so much debate because there
 are
> > > so many DBAs out there who made the wrong decision, and don't like to
 admit
> > > it. Probably too much effort to fix the balls-up.
> > >
> > > Regards
> > > HJR
> > >
> > >
> > >
> > >
> > >
> > >
> >
> >
> >

I have to admit, I'm still at a loss as to where all the hassle with raw is. In these days of of volume managers and backup products, the hardest thing about raw I'm finding generally is 'drag and drop'...

Cheers
Connor

-- 
===========================================
Connor McDonald
http://www.oracledba.co.uk (mirrored at
http://www.oradba.freeserve.co.uk)

"Early to bed and early to rise, 
 makes a man healthy, wealthy and wise." - some dead guy