Jesper Pedersen wrote:
>> 2) binary transfer - still needs some more work and testing. Will wait for 8.3.
>
> Please include in 8.2 driver default on -- or make a 550 release with it in.
> 8.3 is too long to wait for this kind of performance increase.
>
This makes some rather fundamental changes to how data is received and
given the complexity of some of these types (timestamps), this is not
something we want to rush into. Also the usage pattern that enables
binary transfer makes testing and debugging harder because a fetch can
work perfectly on the first attempt because it uses the text format and
only on the second fetch does it go wrong.
The actual performance increase depends on a specific usage pattern that
will not apply to many users. Statement pooling or an extra network
roundtrip to determine the result type prior to execute will be needed
to make full use of binary transfer. I wonder what percentage of users
will get gains from the current implementation?
Kris Jurka