check the data on osadl.org ... with RT-Preempt you should be able to get worstcase jitters of less than 50 us ... or you have a 'bad' system / bad drivers.
With a 'good' RT-Preempt system you get < 20 us as worstcase. osadl is good since they are really hammering the systems while measuring. / regards, Lars Segerlund. 2013/5/6 Kent A. Reed <[email protected]>: > On 5/6/2013 8:13 AM, Charles Steinkuehler wrote: >> Yes, but if you've got a Mesa card, you don't gain that much on an x86 >> platform. Unless you're really pushing your servo rate, rtai and >> xenomai are IMHO really only necessary on x86 if you're trying to do >> software stepgen. But make sure you test your board, latency numbers >> for rt_preempt can vary a*LOT* based on BIOS settings and your >> particular hardware (since for rt_preempt to work well, the driver >> code needs to be written to work well under SMP). > > and, just prior, Ebo wrote: > >> There are people poking at RT-PREEMPT, and depending on your exact >> setup getting very good results. Like others have said, RTAI and >> Xenomai get better latencies and will probably always do so because of >> technical reasons of their implementation. That being said, if the >> rt-preempt latencies is good enough for your application I personally go >> for that -- rt-preempt is now part of the stock kernel and there is a >> variant which is now supposed to provide hard real-time. Anyway, them's >> my thoughts on the matter. >> > > and, just prior to that, Michael wrote: > >> Yes, RTAI and Xenomai have significantly better latency behaviour than >> RT-PREEMPT. I think this is relevant mostly for software stepgen setups > > The question of what is good enough has intrigued me since I first > started reading LinuxCNC nee EMC2 documentation. For many years we have > had essentially only one quantitative criterion. IIRC correctly it shows > up several places but this particular quote comes from the Latency-Test > page on the Wiki: > > " If the numbers are 100 uS or more (100,000 nanoseconds), then the PC > is not a good candidate for software stepping. Numbers over 1 > millisecond (1,000,000 nanoseconds) mean the PC is not a good candidate > for LinuxCNC, regardless of whether you use software stepping or not." > > Is this still the best we can do for guidance to prospective users? I > realize that details matter so it's difficult to get very specific about > what is "good enough for your application" but I keep hoping we could > say more. For that matter, we could say more about the impact of high > jitter. Currently, we say only : "...your maximum step rate might be a > bit disappointing...." If I thought I had the knowledge I would write > more about the impact myself, but I don't and I haven't. MIchael and I > briefly discussed the subject a year ago in the context of stepper > drives and I realized after considerable Internet searching that little > useful information is available (one limited calculation by Proctor et > al and some postings by Mariss) so maybe it's just too hard a subject? > > As a practical matter, we haven't collected much latency/jitter data for > the Xenomai and RT-PREEMPT kernels on different CPU-motherboard-BIOS > combinations. It feels like it is time to start gathering such data in > new tables on the Latency-test page on the Wiki. If I get time later > today (and if the Wiki is more responsive than it was yesterday), I'll > copy in my Xenomai results for several systems and Charles' RT-PREEMPT > results for several other systems which have been posted elsewhere. > Maybe the presence of new tables will embolden others to contribute > their results. > > > Regards, > Kent > > PS - if new material on this subject already has been added to the LCNC > documentation please point it out. My blind spots are widening with age. > > ------------------------------------------------------------------------------ > Introducing AppDynamics Lite, a free troubleshooting tool for Java/.NET > Get 100% visibility into your production application - at no cost. > Code-level diagnostics for performance bottlenecks with <2% overhead > Download for free and get started troubleshooting in minutes. > http://p.sf.net/sfu/appdyn_d2d_ap1 > _______________________________________________ > Emc-developers mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/emc-developers ------------------------------------------------------------------------------ Introducing AppDynamics Lite, a free troubleshooting tool for Java/.NET Get 100% visibility into your production application - at no cost. Code-level diagnostics for performance bottlenecks with <2% overhead Download for free and get started troubleshooting in minutes. http://p.sf.net/sfu/appdyn_d2d_ap1 _______________________________________________ Emc-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/emc-developers
