Re: [rtl] Measurement test example

[Oleg Subbotin <oleg@rtlinux.cs.nmt.edu>]

Hello, everybody.

By "switching time" I mean time required to actually "switch" to/from real-time
process. This time include reprogramming the timer to generate next interrupt,
code to enter/exit interrupt handling routine, real-time scheduler, context
switch time. So if timer interrupt is generated at time X then real-time task
gets control in time X+delta, where delta is overhead. So we tried to use this
information to reduce the delay by reprogramming timer a little bit earlier.

Oleg.

On Tue, 21 Jul 1998, Hawkes, Jacob David C - HAWJD001 wrote:

>  pardon my ignorance but what do you mean by "switching time"?
> 
> regards
> jake
> 
> -----Original Message-----
> From: Oleg Subbotin
> To: Joshua Lamorie
> Cc: Real Time Linux
> Sent: 7/21/98 6:17 AM
> Subject: Re: [rtl] Measurement test example
> 
> On Mon, 20 Jul 1998, Joshua Lamorie wrote:
> 
> > Gidday there,
> > 
> > 	What does the measurement test do?  I inserted the rt_process
> module,
> > and saw the output
> > from the monitor
> > 
> > 	Min: 12 Max: 47
> > 
> > 	Then I inserted the rt_irq module and the output became.
> > 
> > 	Min: 13 Max: 2883
> > 
> > 	What are those numbers?  Latency in microseconds?  What do the
> > different modules do?
> > 
> > thanks
> > 
> > Joshua
> 
> Ok. This example shows difference between "expected" switch time and
> actual
> switch time. So numbers show maximum and minimum "miss" delay.
> This example does not produce any valuable information if you insert
> additional
> modules ( or rt tasks with higher priority ). Using the output you can
> tune the
> scheduler to reduce this time. Find "#define ADJUSTMENT 34" in the
> rt_prio_sced.c and change it to 45 ( == 34+12-1 ; NOTE using 46 is not
> recommended - you can have cases when task will scheduled to run
> _before_ its
> time ). recompile example and run it again. 
> The delays should be distributed between 1 (min) and 36(max). 
> 
> Out tests show that average miss time for "tuned" system is ~0 with
> about 
> 3 to 5 percents(or even less) miss time greater then 4.
> 
> All numbers are roughly microseconds.
> 
> Oleg. 
> 
> --- [rtl] ---
> For more information on Real-Time Linux see:
> http://www.rtlinux.org/
> 

--- [rtl] ---
For more information on Real-Time Linux see:
http://www.rtlinux.org/