Re: Clock Watching the Scheduler 2.2.10unip

Harald Koenig (koenig@tat.physik.uni-tuebingen.de)
Tue, 28 Sep 1999 14:26:17 +0200


On Sep 27, Richard B. Johnson wrote:

> > 1) 500 interrupts take 5 seconds. No surprise there. Except

when pressing the shift key jsut after starting `clockwatcher'
it takes only 3.9 seconds: that's the 30 Hz interrupts from the keyboard.

> > 2) Most interrupts take ~5600 cycles to process on a 530 MHz Cel,

same here on a AMD K6/2-400 (DFI main board).

> > and ~2600 on 200 MHz PPro. This is better than FreeBSD
> > 2.2.8 where they take ~5200 on a PPro 200. But `way longer

> Erm.. How is this supposed to measure interrupts?

it detects/measures just every delay/timeout longer than 90 cycles in this program.
at least some of them have to be from interrupts and their service routines, isn't it ?
playing with differnet interrupt sources shows that these `timeouts' indeed corelate
with interrupt rates...

here is some data from my PC for three tests:

A: free run (X11 etc. running, 83 processes)
B: like A, but shift key pressed after starting, no significant difference.
C: mouse moved pretty fast all the time.

my mouse is a old Logitech serial mouse, running at 9600 baud connected
to a 16450 UART (one interrupt for each character) giving ~430 interrupts/sec
while moving.

in case C the run only takes ~1.1 secs for 500 `interrupts' (makes sense;)
but note the longer time this interrupt service takes (6272 vs. 5632 ticks).

ticks #counts
A B C

5568 1 1 0
5632 22 22 0
5696 7 7 0
5760 4 4 0
5824 5 5 0
5888 2 2 0
5952 4 4 0
6016 5 5 1
6080 6 6 25
6144 6 6 7
6208 2 2 78
6272 2 2 76
6336 0 0 60
6400 6 6 13
6464 0 0 27
6528 1 1 7
6592 1 1 0
6656 3 3 0
6720 0 0 0
6784 4 4 0
6848 1 1 2
6912 3 3 1
6976 0 0 0
7040 3 3 1
7104 0 0 0
7168 1 1 1
7232 0 0 0
7296 1 1 1
7360 0 0 1
7424 0 0 0
7488 7 7 0
7552 5 5 1
7616 12 12 0
7680 10 10 1
7744 8 8 0
7808 4 4 0
7872 5 5 0
7936 2 2 0
8000 0 0 1
8064 0 0 2
8128 1 1 8
8192 1 1 14
8256 0 0 9
8320 2 2 8
8384 3 3 7
8448 1 1 5
8512 1 1 3
8576 7 7 7
8640 1 1 2
8704 2 2 7
8768 0 0 3
8832 0 0 4
8896 0 0 1
8960 0 0 4
9024 0 0 2
9088 0 0 1
9152 0 0 1
9216 0 0 2
9280 1 1 0
9344 0 0 1
9408 1 1 0
9472 0 0 1
9536 0 0 2
9792 0 0 0
9984 0 0 0

Harald

-- 
All SCSI disks will from now on                     ___       _____
be required to send an email notice                0--,|    /OOOOOOO\
24 hours prior to complete hardware failure!      <_/  /  /OOOOOOOOOOO\
                                                    \  \/OOOOOOOOOOOOOOO\
                                                      \ OOOOOOOOOOOOOOOOO|//
Harald Koenig,                                         \/\/\/\/\/\/\/\/\/
Inst.f.Theoret.Astrophysik                              //  /     \\  \
koenig@tat.physik.uni-tuebingen.de                     ^^^^^       ^^^^^

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/