Thread: rtree index very large

rtree index very large

From
Christopher Zach
Date:
Hi,

I've inserted 100000 random boxes into a table (page file size about 8M) and
created a rtree index on it. Indexing needed a few minutes and finally I got
a 200MB page file for the index. Can someone explain this to me?

BTW: I did the same with Oracle 8i Spatial and the index seems to be a
lot smaller. Selection time for geometric queries (non-empty
intersection with a box) is roughly the same - a few seconds for both
2DBMS, but I don't have exact statistics.

I can guess a possible explanation for the smaller index in Oracle:
coordinates are rounded according the given accuracy, which could result
in smaller rtrees, but this is just a guess.

regards,

Christopher

Re: rtree index very large

From
Christopher Zach
Date:
I tried this already (and I did it again to be sure). It had no effect on the
index size (which is ok., since there are probably only few empty pages in
the db).

My initial guess about the reason for the large index is wrong, since the
coordinates in the source data file have just a few digits accuracy.

still searching for an explanation,

Christopher

> try to vacuum database :
> connect on your psql term on your database then
> "VACUUM [VERBOSE] ANALYZE" it. It will clean
> the blank spaces and re-organize the indexes.
> Oracle seems to do it automaticaly.
>
> Christopher Zach wrote:
> > Hi,
> >
> > I've inserted 100000 random boxes into a table (page file size about 8M)
> > and created a rtree index on it. Indexing needed a few minutes and
> > finally I got a 200MB page file for the index. Can someone explain this
> > to me?
> >
> > BTW: I did the same with Oracle 8i Spatial and the index seems to be a
> > lot smaller. Selection time for geometric queries (non-empty
> > intersection with a box) is roughly the same - a few seconds for both
> > 2DBMS, but I don't have exact statistics.
> >
> > I can guess a possible explanation for the smaller index in Oracle:
> > coordinates are rounded according the given accuracy, which could result
> > in smaller rtrees, but this is just a guess.
> >
> > regards,
> >
> > Christopher
> >