Hi,
RE:
> we have a table with many (~0.5 billion) records and a geometry field
> which was defined as a simple "point". The `show table status` shows that
> the row format is dynamic, however, a simple point in the GIS
> representation has a fixed format (see: WKB: 21 bytes: 1 for MSB/LSB, 4
> for type and 2x8 for the two doubles). We experienced that in this case,
> when we include this point field all `select`ions and indexing (alter
> table ... add [spatial] index ...) are much slower rather than if we
> exclude this "point" object.
I made a couple of new tests related to this.
To give further examples, if the table contains a point-type column
(MMpoint) at the end, selection based on another column (e.g. MMI) is
about 2 to 3x slower (with or without indexing on MMI).
Splitting of geometry information from the table to a separate table,
and using join, greatly increases speed, especially if done on
10million or more records.
All in all, it is a great pity that the geometry.point type can not be
chosen as fixed "width".
Gaspar
