Re: [Jack-Devel] new IIO driver added to jack - large overruns

On 22/01/14 03:11, Chris Caudle wrote:
> On Tue, January 21, 2014 1:47 am, Matt Flax wrote:
>> [... xruns description, etc....]
>> This result occurs when I sample 4 channels at 1 MHz.
> That is roughly equivalent to 80 channels at 48k.  Are you sure the
> hardware you have can handle that much real time data processing?

That is a good question. All of the devices stream directly into memory 
using DMA, so my understanding is that there is very little CPU 
interaction at that point. Something must be going on in the IIO 
subsystem to manage the ring buffer, but most likely very little. As the 
CPU appears to be loaded by only 3%, I would say that something else is 
experiencing a bottleneck ... hopefully Christian's suggestion to 
inspect using OProfile will shed some light on what is causing this 
bottleneck.
>> # jackd -R -d iio -p2048
> At 1M samples/second even 2048 is only 2ms.  You will need to do pretty
> careful system tuning to guarantee 2ms response time without any missed
> deadlines.  Do you really need 2ms latency is this system?  That seems
> much lower than what most people seem to have success achieving on a
> general music recording system, even with a -RT kernel.
>
Yes, I understand this point. The problem with these IIO devices is that 
if I increase the size of the ring buffer, and thus am able to increase 
the period size, the xruns actually get worse ! It doesn't seem right to 
me, but that’s what is happening. I have tuned the period size to their 
minimum xrun state.

Matt

1390340677.30109_0.ltw:2,a <52DEEA3A.4070302 at flatmax dot org>