On 9/15/2012 12:22 PM, Jan de Kruyf wrote: > Hallo Kent, > Please let it run overnight, my poor old Dell wanted to go to bed sometime > after midnight and gave some horrible results next morning (been sleeping > at the wheel, obviously). Although it looked quite happy for the first hour > or so. To be sure; I just didn't want to stay awake another minute to get it started:-) > The other thing to watch out for is frequency throttling when not busy. It > bugs your latency straight away. > > https://www.osadl.org/Single-View.111+M52601a5fd69.0.html
Yep. Familiar with the problem from other projects. It's particularly bad with many laptops. I don't think it's the prime mover in my tests so far, but I was killing all the cpufreq related stuff early on. > And the last thing: Some kernels behave better than others on certain > CPU's, and that is probably the nastiest problem of the lot. But this is > life. Absolutely, and this is why I'm spending some time with this board. It already works fine for me in a traditional RTAI-based LinuxCNC system. I decided to play with it because of the general, well-founded belief among our users that Atom boards work great. Inquiring minds (well, at least mine) want to know if this is true with alternative realtime O/Ses. If LinuxCNC over Xenomai or RT PREEMPT Linux is to go forward, the users will need lists of boards that work, possibly more finely classified as to potential use than is our current latency-test results table. > Cheers, > > j. Thanks, Jan. Regards, Kent PS - speaking of old Dells, I recently scrounged a Dell Optiplex 745 (core 2 duo) that I have in mind fitting out with LibreOffice and giving to my grandkids. Obviously I had to test its performance too. It turns out to be a lousy performer with either RTAI or RT PREEMPT. I believe Dell's BIOS is to blame but that's water over the dam. It won't affect kids doing their homework. > > > > > On Sat, Sep 15, 2012 at 5:55 PM, Kent A. Reed <kentallanr...@gmail.com>wrote: > >> Gentle persons: >> >> As a brief summary... >> >> I reset the BIOS on my ASUS AT5NM10-I to nearly default settings, >> including enabled Hyperthreading. I did leave audio processing disabled. >> >> I built and ran the OSADL cycletest (hint: first install the libnuma1 >> and libnuma-dev packages) with essentially the settings they used to >> generate latency histograms except I ran it for 30m to get a feel for >> the test. I get essentially the same results they did. The four threads >> (2cores+hyperthreading) came in at max latencies 42us, 61us, 61us, 53us. >> >> I then ran our LinuxCNC latency-test with a 200us base thread for 30m. >> Two copies of glxgears were running to give a little loading. I actually >> got somewhat better max latencies: ca 35us on base and servo threads. >> >> Conclusions: >> >> 1) the board is probably doing as well as it can with RT PREEMPT and is >> probably useful with hardware pulse-generators. >> >> 2) the latency data on the OSADL website are approximately useful for >> preliminary assessments. In addition to the systems I compared with my >> efforts, there are several systems under test there that (only) >> approximate the boards Charles has tested. The OSADL latencies are >> within a factor of 2 of Charles' results. >> >> Aside: different sites (including our own) use subtly different >> definitions of latency. It would be useful for someone better versed >> than I am to illuminate this topic. >> >> Regards, >> Kent >> >> >> >> ------------------------------------------------------------------------------ >> How fast is your code? >> 3 out of 4 devs don\\\'t know how their code performs in production. >> Find out how slow your code is with AppDynamics Lite. >> http://ad.doubleclick.net/clk;262219672;13503038;z? >> http://info.appdynamics.com/FreeJavaPerformanceDownload.html >> _______________________________________________ >> Emc-developers mailing list >> Emc-developers@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/emc-developers >> > ------------------------------------------------------------------------------ > How fast is your code? > 3 out of 4 devs don\\\'t know how their code performs in production. > Find out how slow your code is with AppDynamics Lite. > http://ad.doubleclick.net/clk;262219672;13503038;z? > http://info.appdynamics.com/FreeJavaPerformanceDownload.html > _______________________________________________ > Emc-developers mailing list > Emc-developers@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/emc-developers ------------------------------------------------------------------------------ How fast is your code? 3 out of 4 devs don\\\'t know how their code performs in production. Find out how slow your code is with AppDynamics Lite. http://ad.doubleclick.net/clk;262219672;13503038;z? http://info.appdynamics.com/FreeJavaPerformanceDownload.html _______________________________________________ Emc-developers mailing list Emc-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-developers
