[Simh] performance of the emulated VAX on different hosts

[A.Bucher at alcatel.de]

The official VUPS benchmark is DEC-internal.  In the 80's and early
90's, Digital Review published a benchmark suite that measured
"MVUPS" ("Microvax II Units of Performance") on a rather wide range of
machines,
with the assumptions that 1 MVUP = 0.9 VUPS or so, justifying the Microvax
aspect with the fact that at that date, there were a whole lot more
Microvax II's
out there than 11/780's and thus it was a more relevant measurement.

The VUP baseline is a 11/780 with a specific (Massbus I believe) set of
disk
drives and memory and a specific oldish version of VMS and compiler.
Running
the same 11/780 with faster disks and newer and better compilers lets you
think that a 11/780 is more than 1 VUP :-).

I think I may have saved a copy of the DR MVUPS benchmark suite somewhere.
Many of the individual tests were floating-point intense, and I have the
gut
feeling that most VAX users today don't care all that much about FP, but
in running it you get all the numbers for each individual benchmark.

Most of us still called them "Vague Units of Performance" :-).

Tim.
_______________________________________________


yep, right.
but since I am not in the benchmarking business, in fact I don't really
care for the real absolute performance, nor do I intend to compare the VAX
with, say, a PC or a zSeries (using their native OSes, of course), etc. The
machine ís as fast as it is. dot. That's it.

I was just thinking about a way of comparing two different host platforms
with the same emulator, to find out what suits besser (like "is it worth to
run simh on a 2.5Ghz PC, or is a 500Mhz enough"). Therefore, a benchmark
would be nice.
Of course, the overall performance heavily depends on lots of things, like
(in the case I mentioned) host system OS performance, IO and memory
bandwidth, disk speed, cpu, etc, etc. Maybe also on the C compiler used to
compile SIMH ... whatever.

That was always my impression of VAXes: Relative low peak performance, but
a good bandwidth. Like, you didn't gain much speed by working alone on a
11/750, but you also did not note a big difference if you were working
together with 20 others. (okay, with 50 users, you could note a _slight_
decrease of performance).

BTW, from the emulation point of view, it should be possible to emulate the
hardware with the correct timing, giving the virtual system exactly the
same performance as it's original mate. but since nobody really uses delay
loops (really ?) to do timing on a VAX, the "speed" of the emulation should
not matter - except for the fact that the whole virtual machine is faster
and perfroms better (imagine a Commodore C64 emulator running at 20Mhz
instead of 1 ... !).

Therefore, I think a go for any benchmark that generates several different
kind of system load, and measure that for comparison.

regards,
Andreas