Re: Re: BDI and servos

>    From: Bob Simon <bsimon@...>
> Subject: Re: BDI and servos

<s>

> How does EMC handle long streams of cutting data? Can it command
> non-stop motion continuously from the hard drive or just the physical
> memory? I can't seem to find any specifics on how EMC handles long
> programs.

I'm going to answer the large file question and what I imagine to be your
next question as well. The ability to contour large programs depends
upon the PC that you are running it on and the specific motion modules
that you are running. Servo planning is much less demanding than
stepper pulse generation so a servo machine will contour faster with the
same machine. A PC will also do reads and writes faster if it does not
have an ISA card dragging it down but this should not be a big problem at
the comparative snail pace that a machine can run.

I believe that file size in Linux is now up in the terrabyte range so you
shouldn't be able to write a file that is oversize. Getting it into the
interpreter is simply a matter of saying, "run this file." Since we
have multiple processes running at the same time, the interpreter is
connected directly to the file to run. The file is not passed through
the user interface or anything else. I've even run files from a machine
over an ethernet without any problems.

I've got to say that I've never had a problem with contouring even with
my stepper driven Griz. EMC uses both read ahead and look ahead. Read
ahead will work to the size of memory available. Look ahead and command
preparation runs about 200 blocks on what I've tested here. Both of
these form a kind of buffer in the command input stream.

Now for the second answer. I had a good conversation with Tom Kramer the
author of the EMC interpreters a while ago. His suggestion with respect
to maximizing the number of blocks that the EMC could read was to
minimize the code that the interpreter has to read/parse by taking
advantage of the modal nature of many of the g=codes. Put one g1 at the
start of a whole block of moves and don't use line numbers. If one axis
is staying the same, don't include it in that block. Keep comments to
the first block in a whole contour.

So using that advice and an 800 MHz cyrix processor, I'm able to
consistently run and hear moves down as short as 0.004 seconds. That
timing was done by the clock in the EMC debug setup. That's about 250
moves a second on average. I have seen some that take a single motion
loop to complete so that would be in the range of 1000 blocks a second on
my machine. With some additional tweaking I'm sure that we could develop
a contouring machine that would be significantly faster.

Hope this helps.

Ray