#linuxcnc-devel | Logs for 2016-03-25

[01:42:57] <memfrob> well nevermind, compiler issues. fixing boost won't help me, night all.
[02:56:59] <TanayG> Hello, I am Tanay Gahlot, Can you please review my GSOC proposal, https://docs.google.com/document/d/1YYe-BUIadKv5wrMeasJenQM4RIfS4erW_LbV5TlORko/edit?usp=sharing
[07:23:18] <jepler> when *not* running a graphics-intensive game at the same time, overnight latency was much better, just 117us. I'll keep it, it'll be excellent for testing/developing mesa drivers.
[07:27:22] <jepler> .. now doing a run with isolcpus=
[07:38:38] <skunkworks> does that work with rt_preemp?
[07:38:54] <skunkworks> I have not played with that
[07:39:04] <jepler> yes, it is accepted on the kernel commandline anyway
[07:39:14] <jepler> and it changes the default affinity of all userspace tasks to exclude that CPU
[07:39:22] <jepler> so on my system with 8 threads, I used isolcpus=7
[07:39:29] <jepler> and now I'm not at the console so I can't monitor the results
[07:40:20] <skunkworks> neat - I will have to try that. Although most systems that I have played with rt_preempt have been <100us so I have not tweeked them
[07:40:40] <skunkworks> or seconds and I didn't even try to fix them ;)
[08:18:37] <skunkworks> http://www.model-engineer.co.uk/forums/postings.asp?th=116062
[08:47:20] <skunkworks> Hmm - I like the new layout of the docs but now I cannot link directly to say 'user interfaces'
[08:47:25] <skunkworks> or is there a way?
[08:47:51] <JT-Shop> you should be able to... hmm
[08:48:32] <JT-Shop> oh, no you can't link to that
[08:49:06] <JT-Shop> hmmm
[08:51:58] <jepler> skunkworks: you mean that on http://linuxcnc.org/docs/html/ you can't link directly to the 'user interfaces' heading?
[08:52:08] <skunkworks> yes
[08:52:40] <jepler> http://linuxcnc.org/docs/html/#sec3 only works if yo have already expanded everything
[08:53:13] <skunkworks> right
[08:58:05] <archivist> breaking links into out docs is a real pain
[08:59:46] <jthornton> ??
[09:03:56] <archivist> people have to update incoming links after somebody has broken something
[09:04:10] <archivist> like http://wiki.linuxcnc.org/cgi-bin/wiki.pl?action=browse&diff=1&id=Released_2.7.X
[09:06:02] <archivist> the point I was making the other day was not that the wiki was wrong but that the destination had disappeared
[09:08:51] <jthornton> it's impossible to know where every link is on the web much less the wiki :(
[09:09:38] <archivist> therefore dont break docs without good reason
[10:32:56] <pcw_home> skunkworks: didn't you try a Intel Pro 100 with hm2-eth?
[10:34:05] <skunkworks> pcw_home, I did a pci card - a long time ago. It also might have been with rt-net
[10:34:24] <skunkworks> (because it was one of the supported cards)
[10:35:01] <skunkworks> I could not tell you if I tried it with rt_preempt
[10:40:22] <pcw_home> Ahh, I thought I did but it was probably a E1000 card
[13:06:39] <JT-Shop> hmm Glade 3.20 has been released with "tons of bug fixes"
[13:10:37] <seb_kuzminsky> haha sounds like one of our release notes
[13:13:15] <jepler> seb_kuzminsky: origin/jepler/2.7/remove-heuristic-delay-warning -> 2.7 or master?
[14:06:59] <seb_kuzminsky> jepler: 2.7 runs on lucid and newer, lucid has rtai 3.8.1, your commit message says all supported rtai versions report latency spikes by themselves
[14:07:15] <seb_kuzminsky> and of course it doesnt matter for rt-preempt, since the code you removed only runs in kernel space
[14:07:33] <seb_kuzminsky> so it looks fine for 2.7, and a nice cleanup, thanks
[14:09:06] <jepler> seb_kuzminsky: OK will push to 2.7. thanks for your review.
[16:41:18] <jepler> 134us with isolcpus=7, so not an improvement. I wonder if I have to isolate both in the HT pair...
[16:58:05] <skunkworks> usually from my little testing - isolating the ones on the same cache
[17:01:40] <skunkworks> (but that was rtai)
[17:08:57] <seb_kuzminsky> makes sense to me
[17:09:58] <seb_kuzminsky> "lscpu -e" might help here
[17:10:41] <seb_kuzminsky> or equivalently, "hwloc-ls"
[17:12:41] <PCW> not sure if its relevant but I had terrible latency with the 4.4.x kernels (or they crashed) this is on a Core Duo
[17:14:16] <PCW> 4.1.15-rt18-rc1 is the best behaved so far on that system
[17:59:29] <CaptHindsight> skunkworks: did you compare what GRBL actually does by comparing it's motion to Linuxcnc?
[18:05:49] <skunkworks> CaptHindsight: I used linuxcnc to log grbls and mach velocity, position and acceleration.
[18:06:24] <CaptHindsight> skunkworks: did you post the log somewhere?
[18:06:30] <CaptHindsight> it's interesting
[18:07:47] <skunkworks> http://www.cnczone.com/forums/opensource-software/271966-grbl-logging-linuxcnc.html
[18:08:24] <skunkworks> http://wiki.linuxcnc.org/cgi-bin/wiki.pl?NewTrajectoryControl
[18:08:47] <skunkworks> there was a long discussion on the mach yahoo group too
[18:18:07] <skunkworks> https://groups.yahoo.com/neo/groups/mach1mach2cnc/conversations/messages/148006
[18:18:41] <skunkworks> https://groups.yahoo.com/neo/groups/mach1mach2cnc/conversations/messages/143177
[18:19:16] <CaptHindsight> thanks!
[19:09:33] <jepler> whee, hm2_eth works dandy on my new machine (using same intel pci-e gigabit card)
[19:13:42] <jepler> (it's not unexpected:)
[19:14:15] <KGB-linuxcnc> 03Jeff Epler 052.7 88227ae 06linuxcnc Merge remote-tracking branch 'origin/jepler/2.7/remove-heuristic-delay-warning' into 2.7 * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=88227ae
[19:32:54] <seb_kuzminsky> sigh
[19:33:14] <jepler> seb_kuzminsky: ?
[19:59:40] <seb_kuzminsky> i've told you like a million times
[19:59:52] <seb_kuzminsky> dont run out of disk or there will be consequences
[20:10:38] <jepler> buildbot again?
[20:10:44] <jepler> :-/
[20:10:46] <jepler> https://emergent.unpythonic.net/files/sandbox/hostmot2-stepgen-1000ppmloss.png
[20:11:12] <jepler> if you look closely, you can see the glitch in position feedback around the 3rd vertical div
[20:11:25] <jepler> this is with simulated loss of .001 of all packets
[20:12:03] <jepler> in this plot, stepgen is being operated in velocity mode with a constant command
[20:13:31] <jepler> servo period is 500Hz, time waited for packet to arrive is (period/10), which will need to be made a tunable..
[20:23:45] <KGB-linuxcnc> 03Jeff Epler 05jepler/master/eth-packet-loss 8634b13 06linuxcnc 10src/hal/drivers/mesa-hostmot2/hostmot2-lowlevel.h 10src/hal/drivers/mesa-hostmot2/hostmot2.c hm2_eth: add a read deadline field to llio * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=8634b13
[20:23:46] <KGB-linuxcnc> 03Jeff Epler 05jepler/master/eth-packet-loss e741660 06linuxcnc 10src/hal/drivers/mesa-hostmot2/hm2_eth.c 10src/hal/drivers/mesa-hostmot2/hm2_eth.h hm2_eth: use read deadline for queued reads * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=e741660