[Simh] Simh on Windows - long startup delays and long shutdown delays

[Villy Madsen]

I haven't seen too much about this lately - I assume that no one has found the
magic bullet yet...

I think that I might have.

It's kind of early in the testing process, but things are looking good...

I have been tinkering with the idea of re-writing the IO stuff to go directly
against the windows api's.  The newer documentation describes the CREATEFILE api
- which is supposed to replace the OPENFILE etc -  with more functionality.  I
have always thought that the issues with the delay startup (and the massive
number of write IOs that accompany the delay) were related with how windoze
handles file sharing.  The CreateFile api allow us to control how (or if) file
sharing occurs.

For whatever reason, I never bothered to look at the higher level C instructions
- after all fopen doesn't have enough parameters to support any control over
file sharing.

Then a couple of days ago I ran across a comment on the microsoft site that the
fopen instruction had been deprecated - and that one should use the fopen_s
instruction instead - unless one wanted to share the file...

fopen_s doesn't appear to be supported in GCC - so unfortunately it means
compiling it with Visual Studio..

I replaced the fopen statement in sim_fio.c with

FILE *fp;
fopen_s (&fp,file, mode);
return fp;

I have seen one instance of the long write on termination and an instance of the
long write on statup on another system that I was testing this out on.  In each
case, it only happened on a single drive.  Prior to this change, it would
usually happen on both of my drives - at both startup and shutdown.

Side benefit, when it did happen it ran faster - the read/write buffering seems
to be better.

Other side benefit - tweaking the optimizations seems to have bought faster
execution than I had with the mingw compiles version.

On my 1.5ghz 1gb P4 I now get 6+ vups.  I was getting less than 5 before.  On
the 3+ghz celeron I am getting ~ 16 vups.  I haven't tried recompiling it on
that system - it would be interesting to see if VS can get any more out of the
code if it's compiling on the target system.

Still a little early to jump and down - but it sure looks like this change beats
the problem I was having.

Ove Christmas, I am going to see if I can modify my autoshutdown code (that lets
windows shutdown the VAX) so that the processor is halted, the memory image is
written to disk and the process shutdown..

Merry almost Christmas

Villy



Villy Madsen
www.members.shaw.ca/villy.madsen