Re: Locally Managed Tablespaces ... again!!!

<ctcgag_at_hotmail.com> wrote in message
news:20030116134537.384$IO_at_newsreader.com...
> "Howard J. Rogers" <howardjr2000_at_yahoo.com.au> wrote:
> > <ctcgag_at_hotmail.com> wrote in message news:20030116123115.542
> > >
> > > I've seen the table/index read serialization/contention done to death,
> > > but haven't seen much here about table/index write
> > > serialization/contention.
> > >
> > > The only write contention into datafiles should be on DBWR, so the
fact
> > > that the blocks were dirtied in serial should no longer matter when
> > > DBWR gets around to writing them, would it?
> > >
> > > Xho
> >
> > Why? You think a single DBWR suddenly acquires the ability to perform
> > multiple writes simultaneously?
>
> No, I think the DBWR could have that ability, depending on how/if the
> OS/platform supports asynchronous I/O. But I think whether DBWR has this
> ability or not is independent of whether the blocks were dirtied
> in parallel or in serial.
>
> > Or that multiple DBWx's deliberately parallelize (and thus create
> > contention) writes to indexes and tables?
>
> If the index and the table are on the same spindle, deliberation is not
> required for this contention to arise.
>

Precisely. But it is then just as likely to arise between a table and its index as it is between a table and a table, or an IOT and a cluster, or a table and an IOT, or an index and another index... and so on.

> > Like, DBWR has a clue *what*
> > it's writing?
>
> My point exactly.

So follow your point to its inevitable conclusion: contention does not inherently arise between tables and their indexes.

If DBWR doesn't know what its writing, then we can re-write your earlier sentence to read

"If the segment and the other segment are on the same spindle, deliberation is not
required for this contention to arise."

That's a fair re-write, because you've just agreed that DBWR doesn't know what *sort* of segment it's writing, so we shouldn't use words like 'table' and 'index', which are things that only we (humans) know about.

And as re-written, that statement is perfectly true. But that's an argument for lots of spindles, not specifically to house indexes in a separate tablespace for performance reasons. Received on Thu Jan 16 2003 - 13:04:36 CST