Hello everyone,
I have a system with 4 GB RAM and am aiming to
write and execute programs that will use as much as possible of this RAM while
running in 32-bit protected mode (via the GO32 program, linked into my code at
compile time). The main problem has been to get access to all the memory through
DPMI - as you are probably aware, Windows will not allow applications to access
more than a theoretical maximum of 2 GB of the 4 GB address space, with 1.75 GB
being the maximum available in practice. Previously I reported that DPMIONE
could see 3.24 GB of memory while CWSDPMI could only see 2 GB. However, the
figure for DPMIONE is what it says is available, but further testing proved that
the number it reports is wrong. I wrote a test program which basically tests
Malloc to destruction by repeatedly allocating RAM 1 MB at a time until Malloc
fails, then prints the number of megabytes successfully allocated. Now I am
probably doing something wrong with DPMIONE (it is quite tricky to set up), but
it could only provide about 770 MB before running out of RAM in spite of it's
reporting 3.24 GB available. However, my experiments with CWSDPMI have proved
much more fruitful. I was chasing a false trail for a while until I realised
that I had to reboot my system when I made changes to CWSDPMI (using the
CWSPARAM program).
With CWSDPMI's default settings it was able to
allocate 2171 GB of RAM as determined by the memory testing program. This was
the same both under FreeDOS and MS DOS (the ME version). The trick is to
increase the maximum swap file size from 128 MB to 2048 MB. The swap file doesnt
seem to get used (there is enough RAM) but CWSDPMI likes to know that it's
there! After making this change and rebooting the system I was able to allocate
3590 MB under FreeDOS or 3595 MB under MS DOS. This is probably about as good as
it is possible to get, bearing in mind the memory needed by the system for PIC
tables etc.
I found nothing in the documentation that came with
CWSDPMI about settings for systems with more than 2GB of RAM, so this has been a
matter of trial and error changing one setting at a time and testing the effect.
Hopefully my findings will be of use or interest - perhaps something could be
added to the distribution docs about the effect of the swap file
size.
Best Wishes,
Roberto
