Joel Jacobs wrote: > Hi Peter, > I think you meant to say 20% not 120% since you gave the example of > 200ms jitter will trip the error with a 1ms servo thread. > I had a chance to do some further testing. I ran the latency test for > longer - a couple hours and the servo thread jitter just barely > exceeded 14us which is exactly 2% of my servo thread (700us) so I'm > thinking the error is tripping at only 2% jitter - remember I have no > base thread. > Yes, for some time I have thought the error message threshold is set a bit too sensitively. A 10% variation of run time for a real time component is certainly cause for worry, but one should leave enough overhead for the occasional burst of DMA or other activity that slows the CPU down. When these error messages occur, more information on the exact amount of run time variance is put in the /var/log/messages file, easiest to examine recent ones with the dmesg command.
With only a 700 us servo thread, your CPU should not be real heavily loaded, so even a 10 or 20% increase in run time is not likely to overrun the next scheduled execution of the thread. Fortunately, these messages are only notices, and don't affect the machining. > I tried turning off hyper-threading and adding the isolcpus=1 to > grub and the latency numbers about halved and no more delay errors - > BUT - It has severely affected the computers performance so I won't > leave it that way. I would rather just clear the error. > I think 2% jitter on the servo thread is perfectly acceptable. Check the dmesg report, and if it really is just a 2% overrun, go ahead and ignore it. > My > motors are running smooth as silk and performing well. > So if it's supposed to trip at 20% and mine seems to trip at 2% > have I discovered a bug or is this a problem exclusive to my system? > Maybe someone else running a servo only system could try setting their > servo period to 40x max jitter and see if the error is generated. > That would make the error 2.5% - enough to trip if it is in fact > tripping at 2% What really is important is the normal run time for the entire thread vs. the scheduling interval. If there is more than 20% headroom there, I wouldn't worry about it. Jon ------------------------------------------------------------------------------ Enable your software for Intel(R) Active Management Technology to meet the growing manageability and security demands of your customers. Businesses are taking advantage of Intel(R) vPro (TM) technology - will your software be a part of the solution? Download the Intel(R) Manageability Checker today! http://p.sf.net/sfu/intel-dev2devmar _______________________________________________ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
