Hello!

I'm not sure that Ignite is able to detect actual affinity 100% of the
time. So that's why collocate=true may help and that's why distributedJoins
may slow down a perfectly collocated query.

Other than that, I don't think Ignite is going to beat Pg on group by
performance right now, when there's a lot of data to process. Maybe
consider using task-split instead of SQL for such case, you will still need
to scan all data.

Regards,
-- 
Ilya Kasnacheev


ср, 5 июн. 2019 г. в 05:37, David <[email protected]>:

> Hi Ilya,
>
> thx for helping me
>
> the column is collocated. Still I need around ~10+Servers to match one PG
> Server speed for a group by.
>
> What do you mean with use "collocated=true"?
> I thought it is only needed in case you dont have collocated data and still
> want to have full result set.
> and its not needed when data are collocated. Do I understand it wrong?
>
>
> btw side question:
> when using sqlline.sh with the "distributedJoins=true"
> all queries are x-times slower than without
>
> which makes totaly sense in case for non collocated data.
> But it is also the case for collocated data, which makes no sense.
>
> sqlline.sh --verbose=true -u
> jdbc:ignite:thin://127.0.0.1/?distributedJoins=true
>
>
>
> --
> Sent from: http://apache-ignite-users.70518.x6.nabble.com/
>

Reply via email to