Re: parallel_max_servers and the number of sessions involved in a SQL

This is why I asked you about all your parallel parameters and what parameters you were leaving to default.
In 19.3 (for example) if you don't set "processes" then the default number of processes is "80 * CPU_count + 40" - which almost looks like Oracle deciding that it has to have 40 processes for the critical background processes and a maximum of 80 processes per CPU is a sensible limit (for an OLTP system).

In the same vein it's perfectly reasonable for someone in Oracle to decide that if 80 processes per CPU is sensible for "normal" processing then 40 per CPU is equally sensible for the "batch-like" processes of parallel execution. In fact they might be thinking in terms of the impact of 20 batch-like processes per CPU on the assumption that DOP 20 usually gets 40 processes but only 20 of them are likely to be very busy at any one instant.

Regards
Jonathan Lewis

On Sat, 28 Nov 2020 at 17:15, ahmed.fikri_at_t-online.de < ahmed.fikri_at_t-online.de> wrote:

> sorry my bad. Indeed, there is a correlation with the process number and
> the pga (which also logical is):
>
> After setting the pga_aggregate_target to 10M (an extreme value) and
> processes to 1500, I got this correlation:
>
>
>
> cpu (host) n_max
> 1 40
> 2 80
> 3 120
> 4 160
> 5 200
> 6 240
> 7 280
> 8 320
>
>
>
> The question now is where the value 40 comes from. (I'll also try this
> test on 11.2)
>
>
>
>

--
http://www.freelists.org/webpage/oracle-l