Re: IMP/EXP accross platforms: What will be missing?

"Rick Denoire" <100.17706_at_germanynet.de> wrote in message news:92agqv8dhkmftvhjpsekuc3h4fc1337377_at_4ax.com...
> ydias_at_hotmail.com (dias) wrote:
>
> >Hi,
> >
> >Objects are created before the import, no grant problems (step 5).
>
> OK
>
> >Why do you think that FTS will run faster if you import objects serially
?
>
> If importing serially, extents will be allocated so that they will be
> physikally contiguous on the harddisk.

You can't say that. Extents end up wherever the file system you use deems it appropriate to put them. Even the blocks of an extent can't be guaranteed to be contiguous one with another.

> When performing FTS on a
> segment composed of those extents, the harddisk head won't have to
> jump around.

Myth alert!!

> If doing it in parallel, extents from different segments will be mixed
> up as they are written one after the other. A FTS will lead to heavy
> movements of the harddisk head in order to chase the right extents,
> since some of them will be skipped.
>
> I assume that neither an oracle data file nor the segments contained
> therein will be necessarilly written in one undivided chunk of space.
> Oracle has no control about this, it is the operating system who
> decides (raw devices are different, of course). But again, I am not
> absolutely sure about this.

You seem to think that the datafile is the O/S responsibility but that extents are within Oracle's area of expertise. But your argument about data files' contiguity is precisely applicable to extents, too. At the end of the day, stuff ends up where the file system places it.

The idea that extents should be contiguous on disk is exactly what underpinned Oracle's own recommendation in about version 5 to try and get all segments in one extent... and we've been suffering from that nonsense ever since (Compress=Y in exports, anyone??). It was never true then, and it isn't true now.

Regards
HJR
>
> Interesting point. I don't think that I will be able to investigate
> this particular question practically.
>
> Perhaps someone more experienced will contribute his opinion.

Or hers, of course. Received on Tue Nov 04 2003 - 17:29:01 CST