[64studio-devel] [64 Studio] #287: Possible kernel regression with high jackd sample rates in 32-bit version

[Daniel James]

Hi Dick,

> I have just run a comparison of loading using Ardour using various 
> versions on my hardware. I played back a four-channel recording with 
> three effect at 44.1 k and 128 sample buffer, 5.8 milliseconds delay. On 
> DeMudi it used 33% processor and on 64Studio 1.4, kernel 2.6.19 it uses 
> 58% and similar with 2..6.19.

Thanks for running these tests! My best guess is that the DeMuDi kernel 
had some optimisation for 32-bit CPUs that we aren't using in our 32-bit 
build. Either that, or the 2.6 kernel just isn't as efficient as the 2.4 
kernel, assuming you are running 2.4 on DeMuDi - this seems much less 
likely though.

Would it be possible for you to snapshot the CPU load of the processes 
on each system under load, to help us track down where the big 
differences are?

If you run the 'top' command in a terminal, hit shift-P to sort by CPU 
load, and take a screenshot, that would be a start. I bet there are much 
better ways of performance profiling though; perhaps someone on the list 
could enlighten us :-)

> My system is a 1.8GHz Athlon with an nforce 2 chip set and 1G ram.

It's just possible that one of the drivers you rely on is buggy, and 
consuming way more CPU than it should in recent kernels.

Cheers!

Daniel