Re: [HACKERS] Profile of current backend

On Fri, 6 Feb 1998, Thomas G. Lockhart wrote:

> > Not exactly. But to get my application running we have to do something to
> > speed it up. This morning I started my program on a database with 15165
> > records in one table and something like 100 in a second plus 2 in a third.
> > Then my software tries to add records to the first table and for each record
> > added it updates a record in the other two. This doesn't count the select
> > etc.
> >
> > Anyway, the data isn't processed fast enough. I tried adding about 600
> > records which should be done (the time in which the data was send) in 5
> > minutes, but our system needed almost 8 minutes to insert the data. And this
> > doesn't cause Oracle too much trouble.
> >
> > So I guess there's need for some speed-up. :-)
>
> I (and others) had done some benchmarking on simple inserts (6 months ago?) and
> had concluded that the speed was similar to other commercial systems (I was
> comparing against Ingres). I recall getting ~50TPS.
>
> This was all before Bruce did his work on startup and runtime speeds. You
> really think your performance is that far off? You are doing selects on the big
> table before inserting? Do you have indices set up?? Our results were for
> inserts on a heap table, which has the least overhead...

    Just curious, but do you have -F set to disable fsync()?  We
really really should disable that by default :(