> If the index scan is reading a large enough percentage of the rows (and
> depending on the clustering of values), it may be reading enough pages
> that there's no advantage (or even a disadvantage) to using the index.
> This is due to both the reads of the index itself and the fact that it'll
> often be reading the values in the main table (it still needs to get the
> commit info from the table data) in random order rather than sequential
> order which can lose some optimizations the OS often gives to sequential
> reads.
This is a really big lose if your index and table are on the same disk.
Jon
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 1: subscribe and unsubscribe commands go to majordomo@postgresql.org
>
