Nick Hill nick at nickhill.co.uk
Wed Apr 26 00:08:58 BST 2006

Raphael Jacquot wrote:
> Nick Hill wrote:
>> This size data set will easily fit into memory of a modest machine. If 
>> the R-tree representation fits into memory, then it is fast. If it 
>> doesn't, it is monstrously slow.
>> How will the database perform when we get to 10 million or 100 million 
>> points?
> that's true. hoever the test machine has only 128M of ram available
> the postmaster process keeps at using 8% of the total ram

You are not telling me what your test results are. I have published my 
timings which so far prove geometric points on MySQL with R-tree indexes 
as less practical for a large GPX point database than the current 
system. I think we can improve the current b-tree tile select speed by 
an order of magnitude using one-dimensional partitioning either 
manually, or as the new feature in MySQL 5.1.

The geometric types may or may not be more practical for human edited 
points and segments.

I would like to try an R-tree index on two 4 byte columns. (two 4 byte 
co-ordinates can represent any position on the globe to +/-5mm). Such an 
index may prove useful for a GPX point database.

More information about the talk mailing list