#emc-devel | Logs for 2007-04-21

Back
[15:12:41] <SWPadnos_> SWPadnos_ is now known as SWPadnos
[15:24:03] <jmkasunich> darn... I could have sworn that I already installed dapper on that SMP box
[15:24:33] <jmkasunich> but I must not have... its still contaminated with win2kserver
[15:25:09] <cradek> that's a simple matter if you have a CD?
[15:25:29] <jmkasunich> yeah
[15:25:38] <jmkasunich> I don't know the pw for the win anyway
[15:25:56] <cradek> it's easy to crack/reset them if you really care
[15:25:58] <jmkasunich> simple != instantaneous
[15:28:44] <jmkasunich> * jmkasunich burns a dapper CD
[15:29:16] <jmkasunich> (I can never seem to find one when I need it, and they seem to die fast here
[15:29:25] <cradek> fwiw you can have the new kernel/rtai and old both installed
[15:32:26] <jmkasunich> I'm doing some benchmarks... building emc2 from a make clean on this box (native) is 7:15 real, 3:45 user, 0:47 sys
[15:32:46] <jmkasunich> building it on a VM on my Core Duo laptop is 3:42/2:11/1:08
[15:33:02] <jmkasunich> I haven't done the VM on this box yet, waiting for the CD to burn first
[15:34:57] <cradek> is that iwth building docs?
[15:37:02] <cradek> seems like mine are a lot faster, 2-3 mins - but I rarely build anything without distcc and at least two machines
[15:39:25] <jmkasunich> no docs (I think)
[15:39:58] <jmkasunich> right, just --enable-run-in-place
[15:41:10] <cradek> real 1m7.886s
[15:41:36] <jmkasunich> what are you running?
[15:41:56] <cradek> oh maybe that's cached
[15:44:04] <cradek> running it on just one machine without cache now I think
[15:44:15] <cradek> looks a bit slower :-)
[15:45:08] <jmkasunich> looks like an E6600 would be quite a bit faster than my laptop (its running a T2300)
[15:45:39] <cradek> we get those at work now - they sure do seem fast
[15:45:48] <jmkasunich> the 6600 has twice the cache, 2.4GHz instead of 1.66G....
[15:46:03] <cradek> especially when you've got 10 of them compiling for you
[15:46:09] <jmkasunich> I think I'm gonna drop the opteron idea and see what I can do with an E6600
[15:46:27] <jmkasunich> the opterons were leading me to dual socket, and the cost was ratcheting up
[15:46:42] <jmkasunich> I really don't need to have two sockets burning watts either
[15:46:47] <cradek> yeah I'd be tempted to get 4 cores then too
[15:47:16] <jmkasunich> intel has (and will have more) quad core CPUs that will fit in the same socket as the E6600, so....
[15:48:24] <cradek> I wonder if they've finally reached the top end of clock speeds
[15:48:36] <jmkasunich> yes, and started coming back down
[15:49:12] <jmkasunich> P4s were at almost 4GHz, but the newer multi-core systems are getting more performance out of half that frequency
[15:49:18] <cradek> I got real 2m26.212s when building on just my desktop
[15:49:23] <jmkasunich> more per clock (and more cores)
[15:49:26] <jmkasunich> what is your desktop
[15:49:36] <cradek> model name : AMD Sempron(tm) Processor 3300+
[15:49:37] <cradek> stepping : 2
[15:49:37] <cradek> cpu MHz : 2000.062
[15:50:12] <jmkasunich> model name : AMD Athlon(tm) XP 2000+
[15:50:12] <jmkasunich> stepping : 1
[15:50:12] <jmkasunich> cpu MHz : 1666.522
[15:50:29] <jmkasunich> something doesn't make sense
[15:50:31] <cradek> we've seen 4MHz to 4GHz in pretty short time, but I doubt we'll see even 40GHz
[15:50:50] <jmkasunich> are the (running but unloaded) VMs sucking the life out of my machine?
[15:51:18] <jmkasunich> I'll have to suspend them and test again
[15:52:20] <cradek> (but it's foolish to predict much about the future)
[16:35:40] <jmkasunich> bah... I just realized the timed builds on the lappy were sim only and the ones on this box were everything
[16:45:51] <jmkasunich> when you are doing the "check CD for defects" thing at the ubuntu iso boot menu, its supposed to show a progress indicator isn't it?
[16:48:13] <cradek> I think it scrolls text at the bottom, I'm not sure if there's a bar
[16:48:33] <jmkasunich> its just showing "loading", didn't replace the original menu or anything
[16:48:40] <jmkasunich> either a bad cd or a bad drive
[16:48:47] <jmkasunich> I should have checked the cd after I burned it
[16:48:54] <cradek> yuck
[16:49:02] <cradek> I hate CDs
[16:49:27] <jmkasunich> me too
[16:49:42] <jmkasunich> I like to use one in the life of the PC (to install the OS)
[16:50:07] <jmkasunich> I was seriously thinking about building a box without an optical drive, to make it smaller
[16:53:22] <jmkasunich> when you timed that build on your desktop, was that a full build, or sim?
[16:53:32] <cradek> it was a realtime build
[16:53:41] <jmkasunich> ok
[16:53:50] <jmkasunich> I gotta figure out why my machine is so much slower
[16:54:13] <jmkasunich> after I get the install started on the SMP box
[16:57:38] <cradek> rebooting smp - cross your fingers
[16:59:10] <jmkasunich> I hate cds... (more)
[16:59:35] <jmkasunich> [2986725.735838] hdc: media error (bad sector): status=0x51 { DriveReady SeekComplete Error }
[16:59:35] <jmkasunich> [2986725.735849] hdc: media error (bad sector): error=0x34 { AbortedCommand LastFailedSense=0x03 }
[16:59:35] <jmkasunich> [2986725.735854] ide: failed opcode was: unknown
[16:59:35] <jmkasunich> [2986725.735860] end_request: I/O error, dev hdc, sector 905176
[16:59:35] <jmkasunich> [2986725.735866] Buffer I/O error on device hdc, logical block 226294
[16:59:35] <jmkasunich> [
[16:59:55] <jmkasunich> boatload of those when I did "md5sum </dev/hdc on this machine
[17:00:00] <jmkasunich> and now that process seems hung
[17:02:11] <jmkasunich> crap... the md5sum process is hung, and even sudo kill -9 won't kill it
[17:05:08] <cradek> yuck
[17:06:37] <jmkasunich> I get a new error in the kernel log every 4-5 seconds, and the cdrom light is blinking...
[17:07:10] <jmkasunich> you think maybe its just doing a crapload of retries, and it will eventually finish the IO, return control to the process, and realize its been killed?
[17:07:39] <cradek> don't know, sorry
[17:07:55] <cradek> sometimes stuff like that does come back...
[17:08:17] <jmkasunich> I could try killing the shell...
[17:08:40] <jmkasunich> heh, it just came back
[17:08:51] <jmkasunich> it must have heard that threat ;-)
[17:12:38] <cradek> reboot 2...
[17:15:40] <jmkasunich> first one not so good?
[17:17:39] <cradek> no initrd on the first one - fixed
[17:17:49] <cradek> actually running now!
[17:18:43] <jmkasunich> cool
[17:21:44] <cradek> RTH| lat min| ovl min| lat avg| lat max| ovl max| overruns
[17:21:43] <cradek> RTD| -118| -133| 2203| 9190| 12622| 0
[17:21:45] <cradek> yum
[17:21:55] <jmkasunich> how hard did you beat on it?
[17:22:09] <cradek> make -j12, moving windows, find /
[17:22:12] <cradek> glxgears
[17:22:20] <jmkasunich> niiiice
[17:22:59] <cradek> you can also isolate CPUs so they're only used by rtai, which paolo says gets you reliably under 10000
[17:23:16] <cradek> now the question is does emc run
[17:24:40] <jmkasunich> well, I got a good burn this time - cdrom md5sum matches the iso sum
[17:24:46] <jmkasunich> * jmkasunich tries again to boot
[17:26:30] <cradek> the answer IS YES
[17:27:58] <jmkasunich> yay!
[17:28:42] <jmkasunich> diskcheck is running much better this time (I know the disk is good, want to make sure the drive can read the whole thing)
[17:28:51] <jmkasunich> then a quick memtest, and install.
[17:29:16] <cradek> strange, I'm getting following errors on abort
[17:29:37] <jmkasunich> exact same config is before?
[17:29:47] <cradek> yes
[17:29:54] <cradek> all three joints give a following error
[17:30:22] <jmkasunich> all the time?
[17:31:00] <cradek> yes I think so
[17:35:39] <jmkasunich> that kinda rules out multiprocessor race conditions - they wouldn't be that consistent
[17:36:29] <cradek> ouch, pos-cmd goes to nan
[17:36:46] <cradek> which makes f-error go to nan
[17:36:47] <jmkasunich> oh
[17:36:52] <jmkasunich> thats bad
[17:36:54] <cradek> wtf
[17:37:15] <jmkasunich> maybe there is a race, two pieces of code in two CPUs writing to the same float HAL pin?
[17:37:55] <jmkasunich> but the odds of one write happening in the middle of another are astronomical
[17:38:07] <jmkasunich> gotta be something else
[17:38:24] <jmkasunich> fpu save/restore maybe?
[17:38:25] <cradek> I'm a bit at a loss on how to troubleshoot this
[17:38:57] <jmkasunich> FPU isn't normally supported in the kernel
[17:39:04] <cradek> you'd think it would be much more broken
[17:39:09] <jmkasunich> yeah
[17:39:30] <jmkasunich> it behaves until you hit abort?
[17:39:37] <cradek> yes it appears to work great
[17:39:45] <cradek> but I haven't turned on the drives yet
[17:39:50] <jmkasunich> you did things like run the splash screen g-code?
[17:39:56] <cradek> oh hey, that time it aborted fine
[17:39:55] <jmkasunich> (without drives is fine)
[17:40:06] <cradek> yes, splash gcode and flowsnake
[17:40:09] <jmkasunich> oh, wonderfull... intermittent, thats worse
[17:40:45] <cradek> it errors almost every time
[17:41:06] <cradek> halscope doesn't seem to want to trigger every time though
[17:42:00] <cradek> strange! f-error-lim also goes to nan
[17:43:15] <jmkasunich> halscope probably doesn't understand nan
[17:43:25] <cradek> a short test shows nice sounding (even) step pulses
[17:43:38] <jmkasunich> is nan above the trigger threshold, or below? ;-)
[17:43:54] <cradek> well, they're both "up" on the scope display
[17:44:12] <cradek> so, above
[17:44:13] <cradek> :-)
[17:46:41] <cradek> all six of the testsuite tests run correctly, that's a good sign
[19:09:39] <alex_joni> 'lo
[19:17:01] <cradek> hi
[19:23:25] <alex_joni> seen your progress..
[19:23:36] <alex_joni> sure seems nice..
[19:24:30] <cradek> yep
[19:24:36] <cradek> it boots and runs on a UP machine too
[19:25:26] <cradek> if emc would work right, it would be even better
[19:29:31] <cradek> alex_joni: www.linuxcnc.org/experimental if you're curious to try it
[19:45:33] <alex_joni> will do..
[19:45:46] <alex_joni> just need to udnerstand why networking stopped working inside my vmware machine
[21:22:13] <jmk-kayak> only a few more steps....
[21:22:17] <jmk-kayak> gotta fix X
[21:29:04] <jmk-kayak> ctrl-alt-backspace restarts X, right?
[21:29:12] <alex_joni> yes
[21:29:20] <jmk-kayak> do I have to close all X apps first?
[21:29:28] <alex_joni> it will kill everything
[21:29:33] <jmk-kayak> so yes
[21:29:37] <jmk-kayak> ;-)
[21:29:40] <alex_joni> you need to close them if you want to save stuff
[21:30:10] <jmk-kayak> I might as well let synaptic finish installing the (non-rt) SMP kernel, then reboot
[21:32:01] <jmk-kayak> x restart first, biab
[21:37:28] <jmkasunich> ah, 1600x1200... room to breathe
[21:46:28] <jmk-kayak> wohoo - 2 CPUs
[21:49:19] <alex_joni> jmk-kayak: nice
[21:49:30] <alex_joni> I just installed the exp. deb from chris
[21:50:30] <jmkasunich> I'm working my way around to that
[21:50:34] <jmkasunich> (in a couple more hours)
[21:51:49] <alex_joni> cradek: no luck here..
[21:51:52] <alex_joni> it stays black
[21:52:15] <alex_joni> oh, wait.. it finished booting
[21:52:29] <alex_joni> only the graphic while booting is busted
[21:55:24] <alex_joni> pastebin.ca/451865
[21:58:27] <alex_joni> that's inside vmware running on doze
[21:58:33] <alex_joni> but with vanderpool activated
[21:58:51] <alex_joni> jmkasunich: I suggest you make sure your processor will allow for Xen to work
[21:59:09] <jmk-kayak> what processor, the ones in this box?
[21:59:14] <jmk-kayak> they're P3's
[21:59:16] <alex_joni> the one you want to build
[21:59:27] <jmk-kayak> the new box will probalby never run a RT kernel
[21:59:35] <alex_joni> doesn't matter
[21:59:44] <jmk-kayak> what is Xen then?
[21:59:55] <jmk-kayak> (I thought it was Xenomai, for RT)
[22:00:08] <alex_joni> Xen is a way to allow more than one OS to run on the same processor
[22:00:14] <alex_joni> kinda like vmware, but different
[22:00:35] <alex_joni> http://en.wikipedia.org/wiki/Xen
[22:02:39] <jmk-kayak> I'm about 99% decided: Intel E6600 CPU, and Intel D975XBX2KR mobo
[22:03:00] <alex_joni> I have an E6600 at work
[22:03:08] <alex_joni> asus mobo though
[22:03:18] <jmk-kayak> is it nicely fast?
[22:03:24] <alex_joni> yes
[22:03:28] <alex_joni> nicely expensive too :P
[22:03:30] <alex_joni> but great
[22:03:43] <jmk-kayak> is there a compabitility list for Xen (going to their site now)
[22:04:03] <alex_joni> they say that any processor with VT-x should work
[22:04:20] <alex_joni> (and almost all dual core processors from intel have that)
[22:05:39] <jmk-kayak> sounds good
[22:05:51] <alex_joni> the mobo needs to have the VMX flag activated
[22:05:58] <alex_joni> but afaik it's a common thing for desktops to do
[22:06:08] <alex_joni> (it's more problematic on laptops though)
[22:06:26] <jmk-kayak> no laptops for me (except the work one)
[22:06:44] <alex_joni> this one is actually nice (the one I'm on right now)
[22:06:53] <alex_joni> T7200 Core 2 Duo
[22:07:12] <jmk-kayak> Mine is a T2300 core duo
[22:07:31] <jmk-kayak> wow, apb-get build-dep emc2 gets a lot of stuff
[22:07:32] <alex_joni> the 7200 is a bit slower than the 6600
[22:07:38] <alex_joni> but not by much :)
[22:07:47] <alex_joni> yeah, all the docs stuff too
[22:09:34] <jmk-kayak> t7200 vs E6600: 2.0G vs 2.4, same cache, 667 vs 1066 FSB
[22:10:37] <jmk-kayak> cradek: stupid question: when you tested EMC with the new kernel, that was a copy you built with the new kernel, right? not an installed version?
[22:12:17] <jmk-kayak> farg... build-dep is downloading the magma UP kernel
[22:15:13] <alex_joni> jmk-kayak: yeah
[22:15:22] <alex_joni> I'm just building emc2 now
[22:15:32] <alex_joni> but configure worked great.. found everything as it should
[22:15:48] <alex_joni> even if I have another kernel&RT installed on the same machine
[22:16:06] <jmk-kayak> it builds for the running kernel, right?
[22:16:32] <alex_joni> right
[22:16:44] <alex_joni> unless you specify something else in the configure phase
[22:16:48] <jmk-kayak> right
[22:17:05] <jmk-kayak> yuck, I'm gonna have a time problem
[22:17:25] <jmk-kayak> stupid installer asked me for my time and timezone... then it assumed the time I gave it was UTC
[22:17:32] <jmk-kayak> result: the clock is 4 hours fast
[22:17:56] <jmk-kayak> and everything thats getting installed right now will be 4 hrs newer than anything that I build
[22:20:50] <alex_joni> clean compile: 2m11.838s real, 0m30.435s user, 1m30.568s sys
[22:20:59] <alex_joni> that's inside the vmware
[22:21:10] <jmkasunich> odd that sys is so much higher than user
[22:21:20] <alex_joni> yes..
[22:21:32] <jmkasunich> my vm tests show that sys is higher than when you run it native, but nowhere near like that
[22:21:37] <alex_joni> I saw things like generating docs taking up a lot of time
[22:21:48] <jmkasunich> I didn't do docs
[22:21:53] <jmkasunich> you are running a fast computer
[22:21:55] <alex_joni> I mean the comp manpages
[22:22:01] <alex_joni> jmkasunich: a fast laptop
[22:22:07] <alex_joni> but the E6600 is faster
[22:22:14] <jmkasunich> my build times were 3+ minutes running native with all the VMs shut down
[22:22:14] <alex_joni> (if that doesn't convince you..)
[22:22:30] <jmkasunich> if they were running (even unloaded) it went to 6-7 mins (memory I think)
[22:22:41] <jmkasunich> and inside the VM was 7 mins as well
[22:23:18] <alex_joni> one of these days I'll have to install ubuntu on this box
[22:24:09] <alex_joni> oh, darn :D
[22:24:21] <alex_joni> forgot I really hacked the emc2 inside there
[22:24:44] <alex_joni> int(inifile.find("KINS", "JOINTS"))
[22:30:42] <jmk-kayak> fsck
[22:30:48] <jmk-kayak> I fixed the clock
[22:30:57] <jmk-kayak> now sudo won't let me in
[22:31:35] <jmk-kayak> I think I'm screwed for 4 hours
[22:31:58] <jmk-kayak> (need CVS to get emc, need sudo to install cvs)
[22:32:03] <jmk-kayak> I suppose its dinnertime anyway
[22:33:00] <alex_joni> heh
[22:34:04] <jmkasunich> gonna be gone for a couple hours, so goodnight (if you aren't here when I get back)
[22:34:14] <alex_joni> good night
[22:34:20] <alex_joni> I just finished the new checkout
[22:41:17] <alex_joni> yay.. it runs
[22:45:30] <alex_joni> haha
[22:45:35] <alex_joni> a bit of overruns
[22:46:56] <alex_joni> pastebin.ca/451945
[23:37:31] <jmkasunich> thats in a VM isn't it?
[23:41:26] <alex_joni> yeah..
[23:41:29] <alex_joni> can you tell?
[23:41:32] <alex_joni> (lol)
[23:42:09] <jmkasunich> well, I'm committed to intel now - bought the mobo
[23:42:21] <jmkasunich> (a local store had it, $10 less than newegg)
[23:42:54] <jmkasunich> its definitely different having a manual and everything
[23:48:05] <alex_joni> I think it's a good choice
[23:50:25] <jmkasunich> two pieces done... just need a pile of others
[23:50:55] <jmkasunich> CPU, RAM, power supply, case (maybe)
[23:58:10] <cradek> alex_joni: I also noticed somethign is wrong with bootsplash/vesafb here
[23:58:24] <cradek> didn't worry about it yet - just used text mode
[23:58:40] <jmkasunich> bootsplash is evil anyway