EBo,

Am 06.05.2013 um 19:44 schrieb EBo <[email protected]>:

> On May 6 2013 8:06 AM, Kent A. Reed wrote:
>> On 5/6/2013 9:16 AM, Lars Segerlund wrote:
>>>  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.
>>> 
>> 
>> Lars:
>> 
>> I'm not looking to open a discussion about the definition of jitter 
>> and
>> the appropriate methodology for measuring it (we've had some of those 
>> in
>> the past on this list, and some of us are quite familiar with OSADL).
>> 
>> What I'm looking for is better guidance to our CNC users, most of 
>> whom
>> find the details about latency as understandable as details about the
>> fuel-injection algorithm used in their car's computer. What we have 
>> seen
>> in the time I've been reading the emc2- lists amounts to constant
>> schoolyard taunting, "my latency is better than your latency." Look 
>> how
>> excited we get when the latest Atom board yields 1us lower or higher
>> jitter results than another. If the better guidance includes pointing 
>> to
>> the OSADL as another source of data (along with suggestions on how to
>> interpret those data) so much the better for the folks who wish to 
>> use
>> RT-PREEMPT enabled kernels (and my thanks to those on this list who 
>> have
>> been making it possible for the average user to even consider using 
>> them).
> 
> Kent,
> 
> What I would like to see, and will help generate as my limited time 
> allows, generate some basic formulas to help guide people.  This is off 
> the top of my head, and is likely wrong given that I am spacey as hell 
> with allergies...
> 
> max_pulse_rate (in Hz) = 1.0 /(maxjitter * Const)

from a user perspective I think measures like latency or perceived maximum 
pulse rates are useless as guidance, for the simple reason that they do not 
express what a user might be interested in, but happen to be some low-level 
number which happens to be measurable easily

what _I_ would be interested in isnt what happens to be easily measurable, but 
what a given setup can do for me

for example:

given a configuration, what is the quality of tracking the commanded path, and 
the speed/speed deviation at which that can be done

tracking quality (I just made that term up) might translate into either a 
maximum path deviation (hard limit), or a distribution with percentiles plus a 
boundary

this probably has both a spatial and a temporal dimension

example for spacial accuracy: for test path X, results are within A uM 95% of 
the time with a maximum deviation of B uM

I admit this probably cant be expressed as easily - I'm lacking theory and 
literature know how here but I would bet there is actually work in the area (or 
in related weapons or rocket research ;)

maybe even there is a way to compute such values on the fly and mirror them in 
(a) pin(s); again that is blue sky and pure speculation


from that perspective, messages like 'unexpected realtime delay' are about as 
useful as dashboard message in my car saying 'cylinder #3 took 0.7mS too long 
to fire':

very interesting detail, now what bearing has that on my goals - do I need to 
stop, can I continue driving or will I bump into another car?

---

at times it might be useful to lean back and ask yourself: does this make sense 
or can we do better than that

- Michael


> 
> where Const is probably something from 3 to 10 and denotes some 
> fraction of base time period that you can put up with the jitter.  
> Knowing how fast you can run the stepgen thread (assuming stepgen) will 
> then allow you to calculate your max speed using a given piece of 
> mechanics/electronics/etc.  For example, if you are using something like 
> the Gecko G540, and motors that can do just over 300RPM, and typically 
> have 200 steps/rev.  We know that the G540 is hard-wired to 10 
> microsteps/step.  So, we have 300*200*10 maximum steps/rev that the 
> motors can do, and 10KHz for the step/sec.  So if we back calculate 
> that, we are talking 100us for the max step rate, and you want your 
> jitter to be a very small portion of that (for argument lets say 1/5 or 
> <20%).  So, if we have a latency of <20us we can easily keep up with the 
> G540.  If I am not mistaken, the latest RT_PREEMPT is able to do that on 
> all my hardware.  Not sure about yours.
> 
> Sometime back I remember reading a paper that looked at the %jitter to 
> duty-cycle and the overall effect on power loss.  I do not have an 
> intuitive feel for that or what implies for the current discussion but 
> this is a start.  I am sure that others will tare this apart but so be 
> it.  This is intended to start a reasoned discussion on spec'ing the max 
> speed of a setup given the jitter and other considerations.
> 
>   EBo --
> 
> ------------------------------------------------------------------------------
> 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


------------------------------------------------------------------------------
Learn Graph Databases - Download FREE O'Reilly Book
"Graph Databases" is the definitive new guide to graph databases and 
their applications. This 200-page book is written by three acclaimed 
leaders in the field. The early access version is available now. 
Download your free book today! http://p.sf.net/sfu/neotech_d2d_may
_______________________________________________
Emc-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/emc-developers

Reply via email to