On 6/7/2015 10:29 AM, Jonathan Cameron wrote:

On 18/05/15 22:17, Sebastian Andrzej Siewior wrote:

quoted

* Xander Huff | 2015-05-14 17:45:04 [-0500]:

quoted

With no other processes running, I got the following results after a
couple of hours on one of our devices:

admin@Xander-roboRIO:~# cyclictest -S -m -p 98
# /dev/cpu_dma_latency set to 0us
policy: fifo: loadavg: 0.01 0.07 0.12 1/176 1473

T: 0 ( 1373) P:98 I:1000 C:6503872 Min:      9 Act:   13 Avg:   13 Max:      51
T: 1 ( 1374) P:98 I:1500 C:4335914 Min:      9 Act:   12 Avg:   13 Max:      49

With a VI reading all default handles (raw, offset, scale,
sampling_frequency) in /sys/bus/iio/devices/iio:device0 constantly in
a while loop, I got the following results after a couple hours on the
same device:

admin@Xander-roboRIO:~# cyclictest -S -m -p 98
# /dev/cpu_dma_latency set to 0us
policy: fifo: loadavg: 6.93 7.30 7.47 3/182 1530

T: 0 ( 1487) P:98 I:1000 C:4497008 Min:     11 Act:   20 Avg:   21 Max:      69
T: 1 ( 1488) P:98 I:1500 C:2998005 Min:     11 Act:   20 Avg:   22 Max:      59

So there is an increase. And there is even a for-loop and I don't know
how deep it is nested. Anyway, do you think it is worth it or would it
be better to get rid of the raw-locks and simply push everything into
threaded context?

Certainly seems likely to be a better way forward to me but I don't
really mind either way.

J

We'll be giving Sebastian's suggestion a try, as time allows.

-- 
Xander Huff
Staff Software Engineer
National Instruments