Re: shared_buffers 8GB maximum

> Not necessarily - it depends on exactly what was changed ... which
> unfortunately I don't know for certain.
>
> Any filesystem call is a kernel transition.  That's a Meltdown issue.
> Meltdown can be avoided by using trampoline functions to call the
> (real) kernel functions and isolating each trampoline so that no other
> code immediately follows it.  This wastes some memory but there is
> very little added time cost.
>
>
> Spectre is about snooping within the user space of a single process -
> it has nothing to do with kernel calls.  The issues with Spectre are
> things like untrusted code breaking out of "sandboxes", snooping on
> password handling or encryption, etc.
>
> Fixing Spectre requires purposefully limiting speculative execution of
> code and can significantly affect performance.  But the effects are
> situation dependent.
>

I don't know the details either. But one of proposed fixes was to flush 
CPU caches after doing system calls. That's the reason why I'm asking.


> So now you know that 32GB is better for your workload than 8GB.  But
> that is not necessarily a reason immediately to go crazy with it.  Try
> increasing it gradually - e.g., adding 16GB at a time - and see if the
> additional shared space provides any real benefit.


That's what we're going to do. Thanks!

Regards,
Vitaliy