Other Directory Sites: SeekWonder | Directory Owners Forum
The following archive was created by hippie-mail 7.98617-22 on Fri Aug 21 1998 - 17:18:38 ADT
Kriston J. Rehberg (kriston@ibm.net)
Thu, 18 Jun 1998 01:23:14 -0400
I am curious if anyone knew excatly what DirectX does to memory when it
starts up a game. It appears to me, at least after playing a few games
written for DirectX for any length of time on a 64 MB machine, that the
DirectX code seems to forcefully swap out anything not having to do with
game activity. I don't believe this swapping to be a memory leak because
the swap file's size gets to be the same size practically every time (about
78 megabytes on a 64 MB system when running, say, Unreal, Quake 2, or
Forsaken), and I seem to remember that one of the purposes of DirectX was to
make Windows latency a non-issue for gaming and one of the side-effects of
doing this might be to swap out non-essential OS services while the game is
in progress.
Would this "forced swapping," if it existed, affect the non-pagable device
drivers that you are proposing?
Kris
Chris Wedgwood wrote:
>
> On Wed, Jun 17, 1998 at 09:12:50PM +0000, Simon R Knight wrote:
>
> > Do you know if the situation is different with page or segment locks?
>
> AFAIK there is presently no way using the Win16/Win32 APIs to mark memory as
> non-pagable.
>
> It certainly does seem possible within a VxD under Windows 95, although I
> never actually managed to see whether or not I could access this memory from
> an application.
>
> I'm not so sure about NT device drivers. I had a look at these back in the
> 3.51 days and printed out about 5 trees worth of documentation before
> getting bored and doing something else.
>
> > I have considered doing this, although I was hoping to discover a
> > pre-existing analysis of this kind. As there are an increasing number of
> > cryptographic implementations/libraries available for the Intel processor,
> > I imagined that this issue must have been treated in some depth, although
> > I have encountered no papers to support this.
>
> An analysis of this might appear at a later date. I had some test code and
> stuff which might be worth re-visiting (if I can find it) and re-working
> with some of the suggestions made by various people on how to minimise the
> risks of data being spuriously paged out. I also never tested NT 5 or
> Windows 98, which merit testing although I assume the situation is no
> different that for '95 and 'NT.
>
> This is actually a wider issue than data just being pages out, Peter Gutmann
> wrote a paper on data erasure/data persistence which makes for good reading.
> Check http://www.cs.auckland.ac.nz/~pgut001/pubs/secure_del.html for more
> details.
>
> > I can devise a series of procedures that will monitor what actually occurs
> > in memory under the different Windows operating systems and modes, but if
> > a paper (or papers) already exist in the public domain, then I would
> > prefer to study these first.
>
> Monitoring the system probably isn't good enough. In fact, its likely doomed
> to fail under some circumstances, mainly because any user can set their
> process priority to be real time (or very near so for NT4 SP1 and later).
>
> Because of this, its not impossible for an ActiveX control, application or
> whatever to do so, then consume vast amounts of ram before your monitoring
> code can detect this.
>
> Of course, it also means your application could spawn a thread at real-time
> priority that does something like:
>
> volatile char x;
>
> for(;!app_exit;){
> /* touch important pages */
> for(m=page_head;m;m=m->next)
> x = m->addr[0];
>
> /* sleep - but not for long, here 10ms */
> usleep(10 * 1000);
> }
>
> but presumably this isn't fool proof, it only shrinks the window of
> opportunity. You would also have to make sure the compiler doesn't optimise
> 'x = m->addr[0]' out.
>
> > Like unix ? Doesn't mlock require superuser privileges under unix ?
>
> Yes, only uid==0 can use mlock, which is a bit of a nuisance in some
> circumstances (like if you want to run PGP or some such of a machine you
> don't have administration permissions on, and therefore can't make it suid
> or whatever).
>
> > As a shareware developer for Windows operating systems, I have no
> > desire to program for another OS.
>
> Just stirring....
>
> -cw
-- Kriston J. Rehberg AOL: Kriston http://kriston.net/
The following archive was created by hippie-mail 7.98617-22 on Fri Aug 21 1998 - 17:18:38 ADT