#linuxcnc-devel | Logs for 2015-03-26

Back
[03:19:42] <archivist> 404 http://sourceforge.net/tracker/?group_id=6744 linked from http://www.linuxcnc.org/index.php/community
[09:23:11] <KGB-linuxcnc> 03Sebastian Kuzminsky 05master 51e3bac 06linuxcnc Merge remote-tracking branch 'origin/2.7' * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=51e3bac
[10:10:32] <Roguish> good morning all. what is the current status of the multicore conundrum? should isolcpus be set in the boot, or what? I'm getting ready to replace my current MB (single core). Does it make sense to go single, dual, or quad core? or does it make any difference at all, and should I go for pure clock speed? I prefer running at least 2.7, if not master.
[10:12:39] <cradek> best possible latency is not really correlated with number of cores
[10:12:49] <cradek> nor clock speed
[10:13:11] <cradek> are you doing marginal software stepping or do you have some other reason to really worry about the latency figures?
[10:13:38] <Roguish> no steppers. only servos with Mesa boards.
[10:13:50] <cradek> ok, what problem are you trying to fix by changing your motherboard?
[10:14:57] <Roguish> well, i keep getting a 'realtime' error, and also just a general upgrade to an old system.
[10:15:42] <cradek> on the wiki there's a list of known-good motherboards
[10:16:15] <cradek> but I prefer to take my live-boot usb stick to goodwill and try a few of the $29 systems and buy one with decent latency
[10:16:34] <Roguish> i've been pouring through the docs (wiki, etc.) and keep seeing problems with SMP.
[10:16:40] <cradek> they have dozens and usually the first one I pick (based on the size of the case) works fine
[10:17:25] <cradek> I don't know of any problem with SMP using the modern wheezy+rtai setup
[10:17:39] <cradek> (the wiki sure can be out of date)
[10:18:45] <cradek> model name : Pentium(R) Dual-Core CPU E6600 @ 3.06GHz
[10:18:52] <cradek> I guess my current mill machine is dual-core
[10:18:56] <cradek> I had never thought about it
[10:19:07] <cradek> it's running wheezy+rtai installed from our livecd
[10:19:19] <Roguish> and, quite honestly, I am unsure of the current status of the development, i.e. realtime kernels, etc.
[10:19:40] <cradek> everything is fine
[10:19:52] <cradek> boot the livecd and it all works
[10:20:54] <Roguish> and that's what I do, generally.
[10:22:40] <Roguish> so, if I purchase a quad core, is there an issue with the SMP, or is that mute now?
[10:23:54] <cradek> I think I have a 12 core behind me. I'll boot the usb stick on it and tell you the result.
[10:24:13] <cradek> (but be aware, I'm sure some quad cores have good and some have bad latency)
[10:25:13] <Roguish> is the SMP issue gone with the current kernels, and the 'new scheduler'? of should I experiment with the isolcpus parameter?
[10:25:52] <cradek> I don't know what issue you're talking about
[10:26:22] <Roguish> http://wiki.linuxcnc.org/cgi-bin/wiki.pl?The_Isolcpus_Boot_Parameter_And_GRUB2
[10:27:01] <cradek> I guess this machine is 8 core
[10:27:15] <cradek> it boots and runs and all 8 cores show up, max jitter is 11775 so far
[10:28:09] <Roguish> am I overworking this?
[10:28:13] <cradek> totally
[10:28:24] <cradek> ideally find a machine you can try before buying it
[10:28:47] <cradek> second best: buy something from the list on the wiki
[10:28:59] <cradek> third best: buy something that's not super weird and try it
[10:29:51] <Roguish> well, I do appreciate your honesty. guess I should relax.
[10:30:04] <cradek> 12170
[10:31:03] <cradek> hey I should latency-test machines from goodwill and then mark them up $200 and sell them to worried linuxcnc users :-)
[10:31:38] <Roguish> good call!!
[10:31:51] <pcw_home> Almost any newer Intel MB is probably fine
[10:31:52] <pcw_home> freeby.mesanet.com/h97-g3258.png
[10:31:52] <cradek> preinstalled even, $300
[10:32:09] <Roguish> yes, i'm chuckling loudly.
[10:32:16] <cradek> this one behind me says "intel desktop board" on it
[10:32:31] <cradek> but it's wayyyyy overkill for running linuxcnc
[10:33:18] <cradek> this one cost me $0 and works great: Pentium(R) Dual-Core CPU E6600 @ 3.06GHz
[10:33:40] <Roguish> so I don't need my Dell T7500 dual 6 core zeon 32g mem. ?
[10:34:04] <Roguish> cradek: you slay me !!!!
[10:34:43] <mozmck> Roguish: if you don't need it, send it to me!
[10:35:06] <cradek> guess that's a 2010 model, which is pretty new for me
[10:35:28] <Roguish> ok, gotta chill and get some work done. thank you gentlemen.
[10:35:38] <cradek> I'm also running linuxcnc (my lathe) on a PIII with 256MB
[10:35:50] <cradek> it runs, but could use more ram
[10:35:57] <cradek> that's what I got the $29 goodwill machine to replace
[10:36:19] <Roguish> hey remember Bill G said we would never need more that 1m ram.
[10:36:30] <cradek> PIIIs tend to have very good latency
[10:36:57] <pcw_home> if you want fanless (and new) the J1800, J1900,J2900 MBs are good
[10:37:02] <cradek> 14157
[10:43:26] <pcw_home> core duos are also good for Preemt-RT:
[10:43:28] <pcw_home> http://freeby.mesanet.com/e8500-preemt-rt.png
[10:44:56] <Roguish> pcw_home: how ya doing the diagram?
[10:45:45] <pcw_home> dgarrs excellent latency-histogram
[10:46:04] <Roguish> where does one find it and how is it run?
[10:46:06] <Roguish> docs?
[10:47:17] <pcw_home> its part of the linuxcnc dist
[10:47:56] <cradek> you can just boot the livecd and type latency-histogram
[10:48:05] <pcw_home> the one in master is nicer (min and max plus stdev)
[10:48:22] <cradek> cool
[10:48:57] <Roguish> got it. COOL.
[10:53:17] <pcw_home> no man page for latency-histogram but --help prints usage info
[10:55:20] <pcw_home> master also has hal-histogram that allows you to plot histograms of hal pins
[12:32:52] <mozmck> c_morley: cmorley1: you around?
[13:11:30] <PCW> Preemt-RT latency is quite good on current Intel MBs also:
[13:11:32] <PCW> http://freeby.mesanet.com/h97-g3258-preemt-rt.png
[13:17:39] <mozmck> That seems too good - are you sure that is accurate?
[13:17:58] <PCW> Yeah
[13:19:20] <PCW> not run very long but RTAI is even better:
[13:19:21] <PCW> http://freeby.mesanet.com/h97-g3258.png
[13:19:38] <mozmck> yeah, but not by much
[13:20:01] <skunkworks> that is impressive..
[13:20:07] <PCW> not much, mainly no "skirt"
[13:20:15] <mozmck> yes
[13:20:31] <skunkworks> I was running the latest patch rt_preempt on the j1900 and it seemed to peak at about 100us
[13:20:32] <mozmck> what model board is that?
[13:20:46] <skunkworks> I was going to try the package manager version for grins
[13:28:05] <PCW> mozmck: which model skunkworks J1900 or the H97?
[13:28:12] <mozmck> H97
[13:29:19] <mozmck> I see this one that looks reasonable: http://www.newegg.com/Product/Product.aspx?Item=N82E16813157547
[13:31:13] <PCW> thats probably fine, this is the one I'm testing with:
[13:31:15] <PCW> http://www.newegg.com/Product/Product.aspx?Item=N82E16813157512&cm_re=h97-_-13-157-512-_-Product
[13:31:26] <mozmck> ok, thanks
[13:32:21] <PCW> mainly because is has both PCI and PCIE
[13:32:35] <PCW> and intel MAC
[13:32:52] <mozmck> is intel MAC better?
[13:34:01] <PCW> well... not sure I though it might be but I still get the best ping times with a lowly PCI $6.00 RTK8139 card
[13:34:49] <mozmck> that's interesting. I thought I had heard the internal ethernet was generally better/faster?
[13:34:57] <mozmck> For Preempt-rt that is
[13:35:01] <PCW> in any case it run hm2_eth reliably at 4 KHz on either the built in mac or a 8139
[13:35:19] <mozmck> good to know.
[13:35:38] <PCW> 8139 being a bit faster for unknown reasons (maybe simpler driver?)
[13:45:11] <PCW> a quick test of ethernet latency under preemt-RT is
[13:45:12] <PCW> sudo chrt 99 ping 10.10.10.10 -i .001 >times
[13:45:57] <PCW> (ignore first ping time )
[13:47:10] <PCW> with the h97/8139/fpgacard I get about 55 usec average
[13:49:11] <PCW> about 66 usec average with the onboard intel MAC
[13:51:04] <PCW> I'll have to try the DC7800, it has a older Intel MAC
[21:46:16] <KGB-linuxcnc> 03Dewey Garrett 052.7 c4c87b9 06linuxcnc 10docs/src/config/images/latency-histogram.png docs: latency-histogram.png fix image mixup * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=c4c87b9