On 04/22/2014 11:17 AM, Simon Riggs wrote:
> They all sound good, apart from the MVCC system catalog changes. I'm
> also not comfortable saying Mat Views again, especially since we said
> it last time and they still don't work automatically in the optimizer,
> nor do they get auto updated.
Actually, we deliberately downplayed the matviews in the last release
because of discussion on this list. Now we have refresh matview
concurrently, which makes them substantially more useful, at least to
me. At some point we need to make them a "feature", and I think this is
that point.
> I think you should add
>
> * Write Scalability improvements - WAL Insert tuning
> * Aggregate performance improvements - tuning & negative transition functions
> * Huge pages
> * Update WAL reduction
What did we do with huge pages?
Do we have any stats for the performance improvements?
On 04/22/2014 11:35 AM, Cédric Villemain wrote:>> "Big features", that
is the ones we want to lead with in press stuff:
>>
>> * JSONB et. al.
>
> we already had JSON (and operator/functions) as big features previously.
You seem to be completely unaware of the new JSON storage format and
features. That's OK; it's why we do this ...
>> * Materialized Views
>
> This was already a big feature for 9.3
Actually, it wasn't; see above.
> * backup throttling (can be linked with time-delayed stdby)
Oh, cool. Can you explain this?
> * GIN improvements
> * WITHIN GROUP
> * huge page usage
> ...
Oh, right WITHIN GROUP. How could I forget.
On 04/22/2014 11:38 AM, Pavel Stehule wrote:
> * support TB in work_mem
I'm not sure about this as a feature. You can declare it, but we can't
actually *use* it effectively, AFAIK.
> * faster numeric sum and avg
> * faster window function - moving average, percentiles
Again, got any figures? People love to know stats when we say "faster
performance" ...
--
Josh Berkus
PostgreSQL Experts Inc.
http://pgexperts.com
