> If Oracle really is doing a sort, it's hard to see where the speed
> difference came from --- unless you have set the tuning parameters such
> that Oracle does the sort all-in-memory whereas Postgres doesn't. Sorts
> that have to go to disk are lots slower.
>
I redid the tests ensuring everybody used 10M sort area... nothing was
significantly altered !! ( altho Postgres moved in towards the big boys on
the first 3 queries and the elapsed time for queries 4 & 5 converged )
>
>
> Hmm, I couldn't make out from your webpage exactly how you did the
> loading, or which steps are included in your timings. I see that you
> used COPY, which is good ... but did you create the indexes before or
> after COPY? What about the constraints? I also see a CLUSTER script
> --- was this used, and if so where is its time counted?
>
> regards, tom lane
My apologies for the state of the scripts ( to all you who downloaded them
for a play) - I had forgotten to complete the README and also left heaps of
test files lying about in the query directory. I have cleaned these up now !
The story is... the comparison was supposed to be simple... so no special
features ( like clustered indexes/tables, bitmap indexes, materialized views,
automatic summary tables...) just a comparison of how well each db did its
"bread and butter" operations.
