Subj : since ya'll are asking.
To : Chris Hoppman
From : mark lewis
Date : Sun Dec 14 2003 05:48 pm
CH> Since ya'll are asking for someone to help.
hehehe...
CH> Maybe you can help me with a problem I am having with
CH> a source code I am modifing.
can only try...
CH> It has timeslices built into it, but the CPU usage is still in
CH> the upper limits (89-98%).
what operating system, version and timeslice style?? all three are needed to be
known... there are several ways of timeslicing... choosing the proper one
depends on the operating system and in some cases, which version of the
operating system...
CH> I sorta remember something I have read and wantted to verify
CH> it. That the timeslices shouldn't be gave at anytime.
i'm not familiar with that...
CH> Like say you do this.
CH> For A := 0 to 100 do
CH> begin
CH> {Do some stuff here}
CH> TimeSlice;
CH> end;
CH> It would do 100 timeslices back to back.
almost... {do some stuff here} would also take some time to do its thing...
CH> I have read that you should time them to be at least 55 ms
CH> apart ( depending on the system ). That if you do like I
CH> did above it will accually make the application use more of
CH> the CPU.
dunno about using more of the cpu or not... and i can't really confirm them to
be at least 55ms apart, either...
CH> Maybe someone could explain to me all about Time Slices. Or
CH> point me in the direction of a place I can read up on them.
wow, i picked up a lot of my knowledge about them in numerous areas several
years ago...
what i ended up doing was to attach my chosen timeslice method to an interrupt
vector (F1 in my case) and then i just called that interrupt when i needed or
wanted to... this let the system chain in the proper timeslice routine based on
the OS and i just called the same interrupt no matter what routine was in it...
my setup uses one of five slicers, depending on the OS... one for OS/2 Warp3,
one for OS/2 Warp4, all other OS/2 and Windows use DPMI slicing... then there
DESQview slicing and finally the old INT28 DOS slicing...
FWIW: the only difference between my Warp3 and Warp4 slicing is the Warp3 stuff
is set to sleep for 2ms whereas the Warp4 stuff is set to 0 (meaning "sleep
now!")...
as for when i call my slicing/sleeping stuff, that depends on my program and
what its doing at the time... if i'm doing a lot of disk processing, i might
sleep every 50 records processed or every 100 lines read... if its pretty much
memory intensive, i may sleep after each second... it really depends... one of
my applications uses both time based and processing based slicing calls...
during intensive disk processing, it uses record based slicing and no timer
based... the reason for this is to keep the effective processing speed of the
program relatively bearable... if i sliced on each record or line read during
file processing, it may take a long long time for the program to finish what it
is doing but it would be very very very multitasker friendly... on the other
hand, if i slice away only once per 1000 records processed, the application
would do the disk processing very fast but be very multitasker unfriendly...
i don't know that i can show the entire code setup that i use because there is
some inline asm as well as numerous asm routines mixed in with the pascal... i
can possibly post the actual slicing calls if desired... i had to play with
them and work some things out for myself... some of my experimenting did cause
lockups and lost data...
