Jeff Bauknecht wrote:
> The problem is context switch time. With Hard real time systems it is <10 microseconds. I am trying to find out context switch times in
> Linux also. There will be several processors and DSP's many pegs many holes
First of all , in my guess, 10 microseconds is really hard!.
One of the solution for this HARD realtime application could be a bufferring scheme.
This idea is simple, add another single chip (like scenex? 100 mips PIC) doing simple processing
and do main processing at 1 millsecond interval. But this is only useful for periodic request.
If some oneshot event (not every 10 micro second) must be served, then you first evaluate
exact how many instructions is necessary for context switching. In my guess you could count
instructions for context switching in RTLinux (with small margin?), because RTlinux has another task scheduler for
real time task and it's flow is rather direct than normal task scheduler (just direct path from interrupt
to higher task, no other intervention from lower priority interrupt). When calculating this, forget caching
and real CPU speed. Cache would not hit because of interrupt, and may memory device (for code/data)
couldnot follow ,say, 200MIPS cpu speed.
Maybe you already know about this, I just want to say that switching time for RTlinux could be
a predictable one.
This message resent by the email@example.com list server http://www.uClinux.com/
This archive was generated by hypermail 2b30 : Sun Apr 07 2002 - 00:01:34 EST