Re: STG2 - Latest Developments

[Jon Elson <jmelson-at-artsci.wustl.edu>]

Will Shackleford wrote:

> As far as the humming is concerned, He suggested removing the rounding that he
> added in January.
>

Great.  I either didn't notice, or ignored the implications of this
change.  I will definitely experiment with it.  It does sound plausible, as
the resolution of my system in X and Y is twice the displayed resolution.

> I tried this on the bridgeport and it had no noticeable effect, but the
> bridgeport was pretty quiet in the first place, atleast compared to the other
> machines in the same shop.

But, my mumbling about the interrupt overload got me thinking about
possible causes for the hanging up of the EMC system when using
versions from, or later than, I think, Mar 2000.  One of the big changes
was to go to a fixed interval scheduling, rather than rescheduling the
RT task every time.  I started thinking that the combination of
a number of things that are scheduled on particular ticks of the
RT task, like the trajectory planning and the GUI.  If all of these things
hit on a particular tick of the 'clock', the RT task obviously will finish,
but one of the other tasks might take so long that it somehow gets
out of sync with the RT task, and causes some problem either to
scheduling in general, or trashes the interrupt structure, or something.
When this 'hangup' ocurrs on my machine, generally the keyboard
goes dead, to the point that there is no way to control XFree86,
like killing it with ctrl/alt/backspace, or getting to one of the console
terminal sessions.  The caps lock key does not affect the caps lock
light.  The mouse goes dead, too, or at least, moving it has no
effect on the screen.  The EMC position display freezes, although
if a program is being executed, it continues moving through the
routine until the end is reached.  If I recall, sometimes an FTP
session can be opened, but not Telnet (but that may be just network
setup details).  With the Mar 2000 or newer EMC, and the 2.0.36
RT patched kernel, this hangup happens anywhere from 10 seconds
to 30 minutes after starting EMC.  With the 20-Dec-2000 EMC,
I have run the system for as much as 12 hours without trouble, and
have never experienced a hangup like that.

I will try turning down the servo update rate to see if it is some sort
of overload, since I am running this on a 100 MHz Pentium classic.

I do want to be able to move up to the newer versions, as the backlash
compensation fix is good, and I don't get that in the Dec version.
There are also a lot of things that Ray Henry is doing that I want to
keep up with.


Thanks very much,

Jon