Re: [boost] RTL

From: Arkadiy Vertleyb (vertleyb_at_[hidden])
Date: 2005-10-17 12:54:27

"Jose" <jmalv04_at_[hidden]> wrote

> Below are my results with g++ -O3
>
> The query should group by city given that in the aslatlong.txt file all AS
> in a city have the same (lat, long). This would make the output more
useful
> (right now the query outputs all the AS from amsterdam first which is
> redundant info given that all AS in the amsterdam area have the same
> lat/long)
>
> ==================================================
> loading table... 21598 records loaded
> 3317 AMSTERDAM ? 991
> [ ... ]
> The query took 11 sec
> 0 records will be added to the index
> 1982 records will be removed from the index
> done with updates
> 5390 AMSTERDAM ? 990
> [ ... ]
> Update and query re-run took 0 sec
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Would it be possible for you to change the code a little, so that it outputs
milliseconds here?

> > Right, so I am afraid we may fall into the trap of comparing the
> > cleverness
> > of hand-coded solutions rather than the libraries' capabilities.
>
> I don't think this is exactly right. Geographic queries should have some
> additional logic that may well be part of the library.

I would rather it belong to a separate library, highly compatible with the
relational one.

> I don't think either
> library has this type of logic now so both might have a hand coded
solution
> for the time being

Right, but the relative impact of these hand-coded solutions maybe too high,
so we may not really be comparing the libraries.

At this point the RTL solution is coded completely as a relational
expression (although rather complex), a single relational view.  I also
might be able to further optimize it inside the relational boundaries.

Please note also that I had to somewhat decrease the performance in exchange
to fast re-runs after the update, so, depending on the context (update and
re-run pattern, etc.), RTL solution might even now be much more efficient
than it looks.

Regards,
Arkadiy