[Info-vax] OpenVMS E9.2 Field Test Begins

[Craig A. Berry]

On 4/8/22 1:11 PM, Arne Vajhøj wrote:
> On 4/8/2022 2:00 PM, Simon Clubley wrote:
>> BTW, talking about performance, I wonder how far away we are from
>> being able to run the same test programs on VMS and Linux on the
>> same hardware and then compare the performance differences.

Probably still quite a ways off.

> It is probably more relevant to compare VMS/x86-64 to VMS/Itanium
> than to Linux/x86-64 for now. The next few years are going to more
> migration from Alpha and Itanium than competing with Linux.
> 
> But to me the logical expectation is that CPU wise C code compiled
> with LLVM backend on VMS will run very similar to the same code
> compiled with LLVM backend on Linux. The different frontends should
> not impact performance much. IO wise we all know that VMS is
> not the fastest.

There have been statements that the cross build environment has
optimizations turned off.  Is that still true?  If so, both the OS
itself and user-written programs just aren't going to be directly
comparable to code compiled with those optimizations turned on.  Even if
it's not true, there may be other considerations.  It isn't clear to me
what impact the GEM-to-LLVM converter will have (if any) on the
optimizations that LLVM normally does.  And things like link-time
optimization probably won't be available on VMS anytime soon but are in
other LLVM-based toolchains.[1]

Meanwhile, the business-level optimizer has squeezed VCI 2.0 and a
couple of different file system replacements off of the roadmap to make
room for the port to x86.  Without those projects, applications on VMS
that do disk I/O or network I/O seem unlikely to be
performance-comparable to similar applications on other systems.  The
only good news here is SSIO, which should make one category of file I/O
more performant since it allows avoiding unpleasant workarounds such as
closing and reopening a file on every I/O.


[1] https://www.llvm.org/docs/LinkTimeOptimization.html