Re: Oracle Myth II - ??

"roobaron" <member_at_dbforums.com> wrote in message news:1791020.1031532588_at_dbforums.com...
>
> Hi Chuck
>
> How is the table mapped to disk?
> Is the tablespace that the table is in have multiple datafiles,
> separated
> over multiple disks?

No, but stiping the tablespace over multiple datafiles would probably imprive the performance even more.

>
> How many controllers do you have for the disks?

One

>
> I am interested in seeing if there is extra performance benefit
> from including the APPEND hint to perform a direct insert as well.

Probably. I didn't use it though.
>
> I humbly asked our Unix guru he thought the best speed you would get
> over 100mb Ethernet is 10Mbyte/sec tops.
>
> So 300Mbyte table copied as a flat file (straight unix copy) should take
> around 30 Secs...
> Interesting to see what the other overhead is...

In a pure network copy he is correct. It should take about 30 seconds. We also have to contend with other network traffic, disk i/o and disk contention on both ends of the link, formatting the data in memory, etc. There are perhaps a dozen other databases running on each server involved in the test all competing for disk and cpu. This was a very unscientific test. None of the other factors were controlled but I am confident that most of them remained pretty static throughout the test.

>
> I would be interested to see the freelist and initrans parameters on the
> table as well.
>
> Good stuff Chuck, the only way to debunk myths to shine the light of
> example on them.
>
> Have Fun
>
> p.s. i can run a test here as well and post the results.
>
> --
> Posted via http://dbforums.com
>
Received on Tue Sep 10 2002 - 07:33:52 CDT