Thread: Max effective number of CPUs that Postgresql can handle?

Max effective number of CPUs that Postgresql can handle?

From
Ron Johnson
Date:
PG 9.6.24 on an ESX VM with nproc=32 and RAM=132GB (We'll be on 14.latest hopefully by February.)

Like the Subject says, is there any point of diminishing returns at which the Postmaster gets "too busy" to manage all the threads?

(I'm not in control of the stack's architecture, so "change _insert_topic_here_" -- including connection pooling -- is not in the cards.)

Thanks

Re: Max effective number of CPUs that Postgresql can handle?

From
Christoph Moench-Tegeder
Date:
## Ron Johnson (ronljohnsonjr@gmail.com):

> Like the Subject says, is there any point of diminishing returns at which
> the Postmaster gets "too busy" to manage all the threads?

It is possible to use 3-digit cores (i.e. 128, maybe more) quite
efficiently. The rest of the system has to fit the amount of compute,
else you end up with an unbalanced system. Also, not every workload
can benefit from this kind of machine. (then: cost of redundancy,
etc.).
I believe there was a benchmark (actual numbers, not just marketing)
done by PostgresPro on a largish Power machine, but I can't find
that right now.

Regards,
Christoph

-- 
Spare Space.