[Info-vax] Some VMS/x86 perf test data from WASD maintainer.

[David Jones]

On Friday, April 28, 2023 at 8:22:29 AM UTC-4, Simon Clubley wrote:
> On 2023-04-28, Volker Halle <volker... at hotmail.com> wrote: 
> > 
> > We can continue the VUPS discussion here, but should discuss the KERNEL mode question in the VSI forum. 
> >
> Why ? The Kernel mode overhead is the real issue that needs discussing, 
> not some artificial CPU speed test.
> Simon. 

I made a wrapper for the prime_sieve.c V2 program (making it a callable function and redirecting its calls to printf)
so I could experiment with making it multi-threaded. My DS10L/466 Alpha can do the 10^9 test in 20.1 seconds if
using a single worker that assumes only 1 worker. If you remove that assumption (create a 2nd thread), the time
goes up to 26.8 seconds due to the ~500,000 calls to pthread_getspecific. My emulated DS20, which is emulating
2 processors, goes from 51.7 seconds to 57.6 for the same experiment. A fair amount of overhead, but multiple
CPUs will improve the final run time.

On my X86 hardware (i5-1235U), the single worker case takes 3.4 seconds while adding in the pthread_calls increases
the run time to 37.6 seconds!  In the first case, monitor shows 100% user mode while the second case show 50%
kernel mode.