Jon Elson wrote: >Stephen Wille Padnos wrote: > > >>If you have a chance to try out a core2duo sometime soon, then try >>starting up a do-nothing CPU hog. I've found that to help latencies by >>roughly an order of magnitude. It's so much better that I actually have >>an embedded application (which starts up at boot time) that does a >>loadusr command that runs this script: >>while true ; do echo "nothing" > /dev/null ; done >> >> >This is hysterically funny! Buy two CPUs, then run a program to >essentially turn one of them off. > >There must be some kind of flaw in the RT scheduling when there >are two available processors, maybe it is trying to "spread the >load" evenly, causing the RT modules to hop back and forth from >one CPU to the other. Maybe that strategy works well for other >environments, but EMC2 is such a small system that it backfires >in our case. > > That's not it (but I don't know what it is). That trick works when you use the isolcpus kernel boot option, which prevents the normal scheduler from using whatever cores you specify. The do-nothing is a userspace app, so that also points me away from the RT code. I should point out that it works regardless of whether I use isolcpus.
What it could be is Intel power management turning off parts of the chip or changing the clock speed (even though I disabled that, I think). >In a remotely related vein, I played around yesterday with >upping the servo rate to 5 KHz on my minimill, and was surprised >at how little of the tuning changed. I did have to pull down I >and D a bit, and the 1/2 F problem was probably worse, as one >would expect by reducing the encoder sampling interval by a >factor of 5. There was a VERY strong 2.5 KHz tome when the >motors were moving at certain speeds, and the PID output had a >clear 2.5 KHz component. I am going to have to experiment with >a simple filter to remove the highest frequency components from >that and see how it works. I think even just averaging the last >two samples will help. > > Interesting. I wonder if a two-tap filter in HAL would fix that. It's a trivial comp to write (if it's not there already). - Steve ------------------------------------------------------------------------- This SF.net email is sponsored by: Splunk Inc. Still grepping through log files to find problems? Stop. Now Search log events and configuration files using AJAX and a browser. Download your FREE copy of Splunk now >> http://get.splunk.com/ _______________________________________________ Emc-developers mailing list Emc-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-developers
