[Info-vax] Future comparison of optimized VSI x86 compilers vs Linux compilers

[Simon Clubley]

On 2020-07-31, John Reagan <xyzzy1959 at gmail.com> wrote:
>
> tl;dr Our memory model isn't one that you can select on Linux and our optimization will be a work-in-progress.
>

When considering memory models and Linux versus VMS performance comparisons
there is one other thing that absolutely needs to be considered and that is
the overhead imposed by the mapping of a KESU based OS to an architecture
which is KU based.

At the moment, no-one (including VSI based on recent comments) knows how
much of an overhead this mapping imposes and how bad things might get
when PCID is not available.

It is also going to be very application dependent and heavy RMS use is
going to impose a higher overhead than something which is compute bound.

As x86-64 VMS moves towards production quality, getting an answer to
this question is going to be important so that we have some numbers
to work with.

The obvious question to ask is if the overhead is going to be enough,
especially when PCID isn't available, to justify moving RMS from
executive mode into kernel mode ?

>From a security point of view, RMS code is effectively kernel mode
code anyway at the moment as you can get from executive mode to kernel
mode without any additional privileges required.

It may turn out that with the help of PCID support this mapping doesn't
turn out to be a problem in real-world applications but at the moment
we simply don't know.

Simon.

-- 
Simon Clubley, clubley at remove_me.eisner.decus.org-Earth.UFP
Walking destinations on a map are further away than they appear.