Mail Archives: djgpp/2003/10/21/10:30:49
> 5) for Charlies Sandman: I will try to use sbrk(), I use to disable swap
> file I will turn it on again (what about performance?).
When using over 500MB of memory with cwsdpmi r5, the page tables will
fault and cause slower performance. If you have a disk cache, or put a
2MB swap file on a ram disk you can lower the performance hit.
But the fix for this is to use the current "test" release of cwsdpmi r6:
http://clio.rice.edu/djgpp/csdpmi6t.zip
I haven't messed with it in almost 2 years (not a lot of requests for more
than 500MB of memory under DOS). Eventually I planned to support 4GB, but
I don't have a system with more than 2GB for testing right now. I am
considering making some major memory management changes for the final
release which would limit the 4KB page space to the first 256MB and use
4MB pages for anything over that. But trying to document that behavior,
which would violate the DPMI specification, is a mess.
> Some question where I will find cwsdpmi r6 and pmode/dj ?
http://clio.rice.edu/djgpp/pmode13b.zip is the most current pmode.
> What about the unreal mode with djgpp executable ?
Not an option. Pmode is the best option for a minimal overhead DPMI
provider, especially if you don't need paging. You might need to remove
HIMEM to work properly in some cases.
> If I use sbrk() I will have performance lost?
No, malloc() calls sbrk(), so you will actually potentially gain
performance. You won't give up extra memory to store the malloc
data structures.
- Raw text -