Re: Some new results

[Tom Jedrzejewicz <tomjedrz@xxxxxxxxx>]

On Wed, 11 Aug 2004 11:36:48 -0500, Joe Pluta
<joepluta@xxxxxxxxxxxxxxxxx> wrote:
> > From: Joe Pluta
> >
> > I was not surprised to find that, as usual, WRITE is ten times faster
> > than INSERT VALUES.  However, I WAS very interested to find that
> INSERT
> > VALUES was actually a tiny bit FASTER than INSERT data structure.  And
> > while INSERT MODS of ten rows was a smidge faster than INSERT VALUES
> > (less than 1% faster), INSERT MODS with 100 rows was actually SLOWER
> > than INSERT MODS with ten rows.  And finally, all four techniques were
> > within 1% of each other, and all were ten times as slow as native I/O.
> 
> I HATE when that happens...
> 
> I should have been suspicious of the results.  I knew it.  While the
> numbers on the INSERT VALUES and INSERT DS are correct, the INSERT MODS
> was bogus.  I was running the wrong program.
> 
> The test results are MUCH more realistic now:
> 
> 1. INSERT 10 rows at a time is three times slower than WRITE
> 2. INSERT 100 rows at a time is 25% faster than WRITE
> 
> So it looks like the break-even point, at least for SQL INSERT vs.
> blocked WRITE, is somewhere around 80 rows or so at a time.  But even at
> more modest 10 rows at a time, it's not a horrible amount of overhead.

Have you tried this with different record lengths and different
numbers, sizes and types of columns?



-- 
Tom Jedrzejewicz
tomjedrz@xxxxxxxxx