Re: [Emc-users] Difficulty defining jitter, was Re: May be of Interest
On Sun, Apr 1, 2012 at 6:15 PM, Jon Elson wrote: > Kent A. Reed wrote: >> >> As an aside, it was interesting to me that while the author of the >> bitmuster article made note of the fact that different tools reported >> different numbers, he/she seemed indifferent to the actual numbers reported. >> >> > Uhhh, why bother making a measurement if you have no idea what > the numbers mean? Yeah, that may make the whole article suspect. > > Jon I understand your thought. But even if we don't know the chemistry behind how the combustion in a car engine works, we can still use it in a practical manner. The same for using some of the numbers about latency, jitter, etc. in tuning our CNC environments. And IMHO, understanding what they mean is easier than understanding the oxidation of organic molicules in a semi-controlled environment (i.e. car engine) :) -- This SF email is sponsosred by: Try Windows Azure free for 90 days Click Here http://p.sf.net/sfu/sfd2d-msazure ___ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
Re: [Emc-users] Difficulty defining jitter, was Re: May be of Interest
Kent A. Reed wrote: > > As an aside, it was interesting to me that while the author of the > bitmuster article made note of the fact that different tools reported > different numbers, he/she seemed indifferent to the actual numbers reported. > > Uhhh, why bother making a measurement if you have no idea what the numbers mean? Yeah, that may make the whole article suspect. Jon -- This SF email is sponsosred by: Try Windows Azure free for 90 days Click Here http://p.sf.net/sfu/sfd2d-msazure ___ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
Re: [Emc-users] Difficulty defining jitter, was Re: May be of Interest
On 4/1/2012 2:03 PM, Jon Elson wrote: > Kent A. Reed wrote: >> With a preempt_RT enabled kernel 2.6.33.7.2-rt30 and an appropriately >> modified EMC2.4.4 [patches from Michael Büsch and Jeff Eppler] running >> on an IBM Thinkpad T40 with a 1500MHz PentiumM cpu, the author measured >> the following: >> - >> 1) the EMC2 latency-test >> >> Servo Thread: 1ms Max Jitter 261101ns >> Base Thread: 50us Max Jitter 101701ns >> >> > Well, these are ghastly numbers! The 1ms thread has jitter of 26% of > the period, and > the 50us thread has jitter of 204% of the thread period! That will > cause system lockups, > I think. > > Jon > I agree, Jon, but that's the different topic I alluded to at the end of my reply. Whether preemption is the right approach or a Thinkpad laptop is the appropriate platform or an otherwise unmodified kernel/configuration is acceptable are all questions that I can't resolve without using the right tools to measure their performance. My point was only that this article clearly illustrates the tools differ even when assessing a single system so we have to be very careful comparing results measured using different tools on different systems. As an aside, it was interesting to me that while the author of the bitmuster article made note of the fact that different tools reported different numbers, he/she seemed indifferent to the actual numbers reported. Regards, Kent -- This SF email is sponsosred by: Try Windows Azure free for 90 days Click Here http://p.sf.net/sfu/sfd2d-msazure ___ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
Re: [Emc-users] Difficulty defining jitter, was Re: May be of Interest
Kent A. Reed wrote: > > With a preempt_RT enabled kernel 2.6.33.7.2-rt30 and an appropriately > modified EMC2.4.4 [patches from Michael Büsch and Jeff Eppler] running > on an IBM Thinkpad T40 with a 1500MHz PentiumM cpu, the author measured > the following: > - > 1) the EMC2 latency-test > > Servo Thread: 1ms Max Jitter 261101ns > Base Thread: 50us Max Jitter 101701ns > > Well, these are ghastly numbers! The 1ms thread has jitter of 26% of the period, and the 50us thread has jitter of 204% of the thread period! That will cause system lockups, I think. Jon -- This SF email is sponsosred by: Try Windows Azure free for 90 days Click Here http://p.sf.net/sfu/sfd2d-msazure ___ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
[Emc-users] Difficulty defining jitter, was Re: May be of Interest
On 3/31/2012 4:45 PM, Kirk Wallace wrote: > I found this: > http://www.bitmuster.org/projects/emc.html > > the thing that comes to mind, considering the rev date, how it seems > fairly significant, and recent questions on the list, it's a little > surprising that this hasn't hit the wiki or been on the lists. Am I > watching the wrong channels? Might the board want to see that this good > stuff gets a some promotion? Are there other things going on that are > worth knowing about? > > I'm not judging or suggesting any action (although just a an e-mail > message with a short title and a link could do wonders:), I'm just > curious and would like to understand a little more if possible. Kirk: Thanks for this. It helped solidify a concern that has been rattling around my brain, to wit, how is jitter being defined and measured in these different activities. Comparing apples and oranges isn't very effective unless one is running a political campaign:-) I read the last section of the bitmuster document and found the following ["EMC2" because of the version involved]: With a preempt_RT enabled kernel 2.6.33.7.2-rt30 and an appropriately modified EMC2.4.4 [patches from Michael Büsch and Jeff Eppler] running on an IBM Thinkpad T40 with a 1500MHz PentiumM cpu, the author measured the following: - 1) the EMC2 latency-test Servo Thread: 1ms Max Jitter 261101ns Base Thread: 50us Max Jitter 101701ns 2) the kernel latency tracer (a built-in) Minimum latency: 1us (nB. This is the first timing bin) Average latency: 9us Maximum latency: 65us <...other details deleted...> 3) the author's "built in latency testing" New maximum latency of task 1 is 112 at period 1000us New maximum latency of task 1 is 113 at period 1000us New maximum latency of task 0 is 99 at period 50us New maximum latency of task 0 is 101 at period 50us - As the author comments, "[I]n my opinion 261us, 65us and 113us are a lot different from each other. Even if you take different measurement methods into account. This is surely some kind of issue." To which I would add, amen. However, as a takeaway, our own latency test appears to be the most conservative of the three, e.g., reports the largest jitter, which means we should be doubly careful about interpreting the OSADL histograms. Note that this is not a judgment on the computer used in the test or the sercos III lashup which was the point of the bitmuster effort. Those are different topics. Regards, Kent -- This SF email is sponsosred by: Try Windows Azure free for 90 days Click Here http://p.sf.net/sfu/sfd2d-msazure ___ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
