hey there jmk
seen lerman around?
I think he forgot to commit a file...
he said something earlier about taking care of some kids, and that he hoped to be able to log on from there
I did a cvs up, and it doesnt compile
yep - he probably forgot to cvs add it again
I can check it in if you like
went to check the farm, and found out that the power supply died :-(
how big a supply, and whayt kind?
"hmm, why hasn't it done a compule since 12/2"
fortunately it is a old server, with redundant supplies
I was only using one, shifted the power cord to the other and its up again
unfortunately that means a replacement won't be simple
ok. presumably there are replacements in the dumpster somewhere? ;)
gotta check ebay
thats one thing I didn't grab extras of
how big are the supplies?
or, what brand of computer?
are they load sharing, or failover types?
presumably it doesn't matter, since you've been running on one this whole time
5V @ 57A, 12V @ 4.5, 12V @ 15.2 -12V at 1.2 and -5V @ 0.5
the system is called "ERS Fault Tolerant II" so I assume failover
oh great - SF is having CVS issues now
there are PS OK lights on the front for both supplies
I may be able to get you a giant 600W dual, non-hotswap, supply from an Antec case
I assume it's AT style, not ATX? ;)
lot of shoehorning to get it to fit
I figure the 900W triple redundant hotswap one has no chance :)
I am a power electronics engineer, so maybe I can repair this one ;-)
get out the smoke detector
(would be easier with a schematic tho)
and look for capacitor juice - that's always good
[01:03:14] <jmkasunich> http://126.96.36.199/support/techinfo/system/ersft2/info/intro.htm
hmm, do I dare??
* jmkasunich rips off the "Warrantee void if removed" tag ;-)
well - unplug it first, and wear gloves
the busted supply is definitely unplugged, I pulled it out of the chassis
(nice chassis, 30 seconds to swap supplies)
I was talking about if I dared to rip the label... ;-)
I was thinking of the search for capacitor juice ;)
under the label is a tamper resistant torx screw
wow - there's someone selling a machine like yours on eBay for $1500
I don't suppose they'd sell me a supply cheap...
they must be smoking something anyway.. they're 200MHz for crap sake
no - this is the 406/33 version
oh wait - this one might not be - there's another auction for a 486DX2/66 unit ;)
work scrapped three chassis and about 20 or more CPU blades
I have about 4-5 spare blades, but I didn't think to snatch a spare supply
of course, for a workstation, a single Athlon 3000 or so would save you enough in power to justify the cost ;)
(or if I did, I forgot where I stashed it)
hard to run five distros at once tho
yes (unless vmware counts)
and the power biss isn't so bad
actually - vmware is the perfect siolution to the compile farm
no high end graphics cards, one HD per slot, no other periphials
how about no video needed, one HD shared by all distros, and a single CPU shared as well
all running at the same time?
slower than native, but still running
and you can do a direct from CD-ROM BDI install?
my Opteron 244 (single CPU at a time) runs Windows 2000 fast enough for SolidWorks (3D CAD) and Altium (high end electronic design)
actually - direct from ISO, no CD needed ;)
the install is "normal", you don't have to hack anything?
the performance in vmware is roughly equivalent to an Athlon XP3000
right - just set up a VM with IDE as the virtualized controller, and installed BDI onto it
also puppy, and windows
can run all of them at once, though things slow down a little
you want to host the compile farm?
(but not compared to a P-200)
that's my main work machine
what about an older one then...
If I can get a spare athlon or something, then I may
the actual load isn't much - each slot does a cvs up once per hour, takes 30 seconds or less unless there has been a truly massive commit
if there was a change, it does (./configure;make clean;make)
I might donate a computer to you for it - I'm not exactly a shell guru, and have no idea how the scripts riun
if you help with the vmware stuff, I'll handle the scripts
I can do that
what are the CPUs in the machine now?
actually, they were written with distributed use in mind
Pentuim 200, 128M ram
shouldn't be too hard to get that level of performance out of a modern processor
anybody could check the scripts out of cvs and with help from me (or a single page of instructions I should write one day) have their box act as a slot
the farm takes 11 to 25 mins to do make all, depending on the distro
that's one reason I want to be able to compile on the Opteron
the fastest is the BDI-2.18 slot
the slowest is BDI-Live
this machine takes 15 minutes to do a full (all modules) kernel compile
how long for emc2 make clean;make?
kernel make clean ; time make bzImage modules = 15 minutes
can't tell, I have no RT on this machine
(I've never done a kernel build, that benchmark is meaningless to me)
I can boot the BDI and check though - hold on
well - it's around 100M of source, if that gives you a comparison to emc ;)
though I shouldn't say it's an "all moodules" build - it's just a rebuild of the stock ubuntu kernel and modules, which support most everything ;)
finally un-tamper-proofed the screws and got the cover off
this is (was?) a quality supply
fat wires eh:?
fuse in a clip, actually replacable without soldering
(and not blown... surprizing)
OK - I have emc2 doing a full make on the BDI virtual machine
no cap juice in sight
jmkasunich: did you ever figure out the compile problem?
the lerman one?
he forgot to cvs add a new file
his mods to the existing file (and the makefile) are there, but the new file isn't
OK - 1 minute, 48.511 seconds
jmkasunich: want me to go fix it?
if you can, I thought lerman would have to do it
the file is still there in the lerman-interp branch
let me go look at it
neat... the supply is locked into the hot-plug slot by a little bracket
which hits a microswitch and shuts it down before you can disengage it
in its day, this was some first class gear
I like working on that kind of stuff
I think the antec has similar features - the thing I didn't like about it is that all 3 supplies are fed from the same IEC inlet
yeah, this one had 2
smarter that way - you can actually use separate UPSes
mine still has a simgle point failure mode
there are 5 installs in the compile farm (at this point), right?
well unless you are in a rather exotic building, they all do - the grid
one is just a router, 4 are doing compiles
no - UPSes would solve that, for some amount of time
I think that a reasonable low level Opteron or A64 machine would outperform the compile farm using vmware for the 4 guest (client) installs
a reasonably fast Athlon XP would likely match it as well.
in practical terms, even a 1G machine would probably surpass it
the slots aren't synchronized, so its rare that all four fire off at once
with only one running at a time, the full power (almost) of the host can do the make quickly
hmmm - I've actually run emc in stepper mode under vmware - maybe I'll look at how bad the pulse outputs are with the scope ;)
of course, when I replace a power supply, then all four start up at once (or close, I login and kick the scripts off manually, never bothered with cron or anything like that
the scripts should be in a cron job - no reason to have them doing their own timing
unless you don't want to depend on cron (or install)
sleep 3600 is easier than learning cron
hmm, mosfets seem fine, no shorts on the primary side of the switcher
can you see (or smell) burnt chips?
no bad smells
no smoke stains
did you check the fuse? ;)
I hate it when I forget the basics
20:23:38] <jmkasunich> fuse in a clip, actually replacable without soldering
[20:23:46] <jmkasunich> (and not blown... surprizing)
and 3 hours later smack myself in the head
ah - misse that
along with the 'd' in missed
I'm wondering if there are some basics here...
IEC socket contacts still connected, ...
so - shrinking the terminal window sped up the compile - it was 1:34.214 for the newly downloadde HEAD
the PS plugs into the huge backplane, and the IEC sockets are connected in back
ah - oK
can't rule out something as dumb as the IEC cord working loose, but I think I checked that kind of thing before I switched over to the backup
are the old BDIs still available? I should just make sure they'll install on this kind of hardware (and software)
I think you can still get 2.18 (or 2.20) and Live RC46 isos
TNG isos got hard to find a while back, but I think I have a copy
I can make you a login on cncgear.com if you'd like to upload
cncgear? thats one of the mirrors, right?
is that you? didn't know that
yep - a hosting service, but mine
actually - I was going to mention them as a viable candidate for linuxcnc.org
I would like to preserve the TNG, even tho Paul has written it off
what was it based on?
lemme check if I still have the isos, might only have em on CD
but you shouldn't drive classics - they lose value
[01:49:37] <jmkasunich> http://www.isw.uni-stuttgart.de/personen/t_franit/echtzeitlinux/download.html
thats where I got them from, seem to still be there
amazingly, I actually saved the URL in the same directory as the isos....
ok - I'll download it
probably because it was a bear to find in the first place
jmkasunich: is that the last bdi capable of running emc1?
BDI-Live will run emc1
that came after TNG
TNG is old
TNG is RedHat 7.2,3
2.x is oldest (based on RH6), then TNG (based on RH7.2) then Live (based on Morphix/Knoppix), and finally 4.x based on Debian
BDI-2.xx is Red Hat 6
ok, got it
and they can all run emc1 except the 4.x family
yay, one slot is reporting a good compile now
the others should finish over the next 10 mins or so
rayh: did you see the changes I checked into halcmd today?
No I didn't get away from customers until just now.
Will get em now.
ok - how'd it go?
Can you describe a bit of what you did with the show
Oh Good. They were tickled with HAL and CL.
headers are gone (no "components" "name FP, etc")
component names are printed instead of ID numbers with show pin, param, and func
there is exactly one format for the value - no extra hex version on the U8 and U16 types
threads are printed one per line - shows the normal info, then the functions in the thread, separated by spaces
same deal with signals
the arrows still print, so each connected pin is two tokens, like "==> ppmc.0.stepgen.00.value"
getting it now.
just do some command using the option -s, you'll see the differences
(but be careful - it's ugly to the human eye ;) )
I spent some thought time on the nodes idea.
actually - to see the differences, do 'bin/halcmd show all > foobar', then 'bin/halcmd -s show all > foobar2', and look at the differences
ok - what did you come up with?
For pins we loop through the pin name
the base of the name ppmc.0 becomes the main node
for that set
then digging down one sub part of the name produces that level of nodes
similar to a directory tree, which gets split at '/', this would be split at '.'
Right. What I want to do for some of the names like we were looking at last night
there were as many as three chunks . separated before we got to any changes
I'd like to make all of that the top node.
we should talk more about the naming conventions though - I don't think we really finished that last night
The exact sort is a bit confused in my head yet.
No we didn't.
That is one thing I'd really like to see happen before the release.
how do you like the idea of just "ppmc.0.dout-00"?
the dout is complete
* rayh takes a minute to look at ppmc pins
right now it's "ppmc.0.dout.00.output" or something like that
ppmc.0.dout.00 does not cause any confusion as far as I'm concerned.
the din signals have both in and invert
and conversely ppmc.0.din-00
and I'd make the invert a parameter, just like the outputs
oh. That took a minute. Sure a param sets the signal polarity.
keeps symmetry with the outputs
invert on inputs shouldn't be a param
you might need both inverted and non-inverted versions
you can always use an inverter block if that's necessary
although depending on the situation, CL might be able to take care of that for you.
ok - and TP discussion ;)
here we go again
you know, I intended to not turn on IRC, and just code tonight
sorry to bow out - I'm not sure I can be of much help there
but the missing file had me looking for lerman, and......
did you see the halcmd changes?
yes, haven't run it, but read the commit message
looks good to me
ok - let me know if you think of any changes to the output
Good job on this stuff guys. I'll go test now.
good. let me know if you need any changes
HAL:14: ERROR: parameter 'ppmc.0.stepgen.00-03.pulse-width' notfound
back so soon? ;)
show param ppmc.0.stepgen, see what the name really is
I'll have to comment these in the hal in order to get that far.
* jmkasunich bites his tongue
Ah. I know the problem. config is pwm, and board is usc
or the other way round.
if the config was pwm, it should be looking for ppmc.0.pwm.something
other way around, yes
you really need two little scripts ray:
Yes I do
start: sudo realtime start; bin/halcmd -kf <yourhalfile>
stop: sudo bin/halcmd unloadrt all; sudo scripts/realtime stop
good, I can re-do those without all the mistakes I made the first time ;-)
start: sudo scripts/realtime start; sudo bin/halcmd -kf <yourhalfile>
that one loads realtime and your hal file
then you can do halcmd show, or whatever you want
the -k will make it keep going after errors
incidentally, I added a prompt for -f (stdin) mode
stop: sudo bin/halcmd unloadrt all; sudo scripts/realtime stop
but left it out in script mode
that one stops and unloads stuff
saw that too, nice
do you think it should be there in script mode as well?
where are you gonna add readline()?
no, not in script mode
we had discussed the fact that it would be a "done" indication
for a piped halcmd
SWPadnos: what was that -s thing again.
bin/halcmd -s show pin ppmc
to get the unreadable version ;)
odd no difference here
must not have gotten the good stuff. will try again
if you run bin/halcmd -h, do you see the -s option in the help?
hal_ppmc bit R- FALSE ppmc.0.stepgen.03.enable
That look like your work?
I made a mistake during transfer and compile. To many phone calls.
so hal_ppmc is the owner of the pin
instead of "03" or whatever the module ID was before
hal_ppmc has a bit, read (by the module), currently false, called ppmc.0.(etc)
I'm not sure I like the way signals are printed at the moment
the arrows are kinda lame ;-)
I had considered walking the list twice, so I could print the outptut pin first, then the input
read my mind again
but if there's no output connected, ...
or no input
so you still need to indicate direction in the print
name !no-driver !no-load
or name driver-name !no-load
or name !no-driver loadname1 loadname2 loadname3
name driver ==> load1 load2 load3
I like the last best
name ==> load(s) means no driver
ok - that should work
name driver ==> \n means no loads
easy to code too
name ==> means nothing
and easy to parse
right - that was the other problem
actually - it's not as easy to parse with script language
they usually work at the word level
though I'm sure it's possible to iterate through the tokens in some fashion
got to bail. started to early today. see you guys tomorrow.
ok - see ya
did just a bit of work with -s and it looks great to me.
SWPadnos_ is now known as SWP_Away