Compile RTAI kernel 3.4-9-rtai-686 on deb.wheezy

To try this optimization i must select a different kernel.
The 3.4.55 haven't Atom optimization. Only a generic x86 optimization. A newer kernel must have.I'll find and try.
Some suggestion on newer
kernel/rtai/config
mix?

Hi

You need to choose a kernel which has a rtai patch in the version of RTAI you have used.
That specific version was used by Seb because there are huge problems with later ones.

I would suggest the 3.5.7 kernel.
There is a patch for it and I use this on my workshop machines.

There is a config_MATOM option that optimises for the in-order pipelining architecture of the Atom.

NB. You may even find the same option in 3.4.55, even if the processors field does not show Atom as a choice.

regards

I'm tryng 4 platform actually
Celeron dual core
Celeron Quad core

Atom dual core (coming soon)
Atom quad core (coming soon)

I'm testing latency of the four cpu (for now 2).
But i have a strange issue.

Celeron Quad core was slower than Celeron dual core.
I read this
wiki.linuxcnc.org/cgi-bin/wiki.pl?The_Is..._Parameter_And_GRUB2
And i can solve the problem disabling 3/4 cpu from motherborad. But sometimes latency goes crazy.

There is a new scheduler in RTAI for 3.10 and 3.14. It no longer requires any isolcpus settings.
3.14 RTAI is in the testing stage at the time of this writing (September 2014). It should be in the official Linuxcnc release in the near future.

Seb Repo is RTAI 3.9 right?

Latency often goes up with the number of cores...

Also the latency test can be rather misleading about real world behavior
I would take a look at the time's and tmax'es of various linuxcnc functions.
You will find that some CPUs that show good latency on the latency test (Atoms for example)
have terrible actual servo thread performance (a D525 has real trouble running a 2 KHz servo thread
since it has ~100 usec latencies _after_ the thread dispatch latency
(and thread dispatch latency is all the latency test measures)

There is a new scheduler in RTAI for 3.10 and 3.14. It no longer requires any isolcpus settings.
3.14 RTAI is in the testing stage at the time of this writing (September 2014). It should be in the official Linuxcnc release in the near future.

Seb Repo is RTAI 3.9 right?

The reason that Seb used that old version of RTAI, is because of the problems with later versions.

The main issues revolve around the maths functions, sincos in particular and also sse.
You have appreciate that the rtai project have no interest in linuxcnc running on it per se, and they broke it for linuxcnc for quite a while.

It was worked around in Machinekit by bringing all the maths functions into an internal library and not using the rtai one.

ShabbyX and NTULINUX have fixed a lot for linuxcnc, but you may find the emphasis has shifted to amd64 platforms and x86 is not well supported in later versions of rtai.

So if you just want to build specialised kernels for the 3.4.9 based distro, then stick with the RTAI used.
If you want to experiment further, then just be aware of the pitfalls you will face.

regards