Re: [Info-Ingres] Generate 1 million rows of unique data - Fridayfun

On Oct 24, 2008, at 10:50 PM, Paul White wrote:

>
> Can someone say why the transaction log fills with over 300Mb of
> stuff?
> And why these two cases perform so differently?
>
> create table millions ... with nojournaling ; \g
> insert into millions select .... ; \g
> 240 seconds
>
> and
>
> create table millions as
> select .... with nojournaling ; \g
> 11 seconds
>

Trace point op149 will explain it. The first uses row-by-row PUT,
the second uses bulk LOAD.

I made a mod at Datallegro to do insert-select with bulk LOAD, but
only for non-journaled heaps. It was a fairly intrusive change so
we haven't propagated it to the main Ingres code yet -- but it's
on the list.

(likewise, create as select with LOAD when the table is journaled
but the database is not.)

Karl