isolcpus=1 not activated ? help ?

Hi,

This is my first post but I did allready read a lot.

I want to setup linuxcnc on a fresh machine and have acceptable latency values.

I use/bought an Intel D525MW board since I read on this forum that this is a board that gives rather acceptable latency.


I did an install of ubuntu 10.04LTS from livecd and disabled Hyperthreading in my bios.


With my system active and about 20 opened glxgears applications, I get rather high latency values reaching upto 30000

However, with only 1 glxgears running fullscreen the latency is allready quite acceptable. (read <15000)


I wanted to improve this further by implementing the isolcpus=1 param to grub.


Now I am absolutely NOT a linux knowing person and I did the following.

1) I had to enter su command to get admin rights
2) with the nano editor I edited the file /etc/default/grub. When I opened the file there was mentioned on line 9

GRUB CMDLINE_LINUX_DEFAULT="quiet splash"

I changed that to

GRUB CMDLINE_LINUX_DEFAULT="quiet splash isolcpus=1"


then I rebooted the machine and tried again, but system monitor (and top also) still shows activity (a lot) on bopth CPUs.


Did I do something wrong ?

Kind regards,

Bart

Hi

Andy Pugh wrote:

I think you also need:
sudo update-grub

To copy the new stuff to the actual grub config file.

Exactly so, the only reason you edit the GRUB config file is because that is persistent, whereas any direct edits of grub.conf would be lost next update

Once done, run cat /proc/cmdline which will show the boot parameters, just to check.

Then to check per CPU usage run mpstat -P ALL ( you may have to install it first)

Then hopefully the next latency test will show improvement

regards

Ok thank you. I'll do that.

I have an extra strange question.

I bought a D525MW motherboard because people report it as GOOD for linuxcnc.

When I run the latency test (or anything else) on the ubuntu 10.04 , then the first thing I did was open the power management app and tell the pc that it should NOT go to sleep after x min of inactivity. I also clicked to make this the default.

Then I went to the screensavers and there I put the slider on 2 hours and also unchecked the line telling that screensaver should be activated. I deliberately selected a screensaver in the selection list that SHOWS something on screen. It is disabled so it shouldn't come up anyway, but if it would, I would immediately recognize it.


Well, when I run the latency test I see that after a couple of minutes (still have to time this) the screen blanks anyway. It is not the screensaver because I would recognize that one I selected.

Are there other kinds of screen blanking built in in ubuntu which I might not be aware of (its a fresh 10.04 install from live cd)

I have the impression that there is still powermanagement active or so. Can I disable some more stuff somewhere?

Kind regards,
Bart

With "xset -q" I discovered that it doesn't make a lot of changes if you change by using the applets.

It still showed me that there was blanking on and dpms was also on.

So I used the following lines

xset -dpms
xset s noblank
xset s off
xset noexpose

now there seems to be no more screenblanking and my latency is also in the range of 15000 to 16000 maximum.

I must admit that I like to test with 10 or more glxgears opened at same time and fullscreen + after that trying to open a firefox, so it might stress the system well.




Now 16000 would not be so bad for my lasercutter I think. The only thing I still don't get is how base period is calculated based on jitter.

The only thing I still don't get is how base period is calculated based on jitter.

This guide lays out the basics of calculation, but a lot is dependent upon getting the right figures for your steppers and drivers
wiki.linuxcnc.org/cgi-bin/wiki.pl?Tweaki...ftwareStepGeneration

As a rough rule of thumb I start with a base period 2 -3 times the max jitter figure. This is considerably higher than the theoretical figure usually.

From there you can reduce it until you start getting following errors on rapid retractions etc.
The figure you end up with can be affected by what acceleration figure you have set, what microstepping rate is selected, as well as the actual velocity.
Many people set their acceleration too high and this causes problems.
Using micro-stepping above 4 - 8x is also often counterproductive, the step generator has to work extremely hard to move the axes anywhere, and these steps are interpolated rather than physical, with inherent risk of over or under run.

You are at the fun part now, good luck