Re: wild I/O interface idea

[Jim Register <jtregister-at-att.net>]

At 02:53 PM 4/27/2000 -0400, Jon Elson wrote:
>
>Doug Fortune wrote:
>
>> Regarding the previous ISA/PCI/USB/ethernet etc
>> conversations on how to get signals out to/from the real world
>> in a timely manner for steppers and servos.
>>
>> The problem with ISA is that it is constrained to
>> operate on an 8MHz cycle.  I believe the accesses to the
>> parallel port and system timers (for RT-Linux) also
>> must pass over this bus.
>
>No, the timer chip (assuming this is the system timer) is
>on a motherboard chip set, and should be at near processor
>speed.  If the parallel port is on the motherboard, the same is
>true.  If the parallel port is a card plugged into the ISA, then it
>will be restricted to ISA speed.  Since practically all Pentium
>motherboards have at least one on-board parallel port, that
>one should do fine for the real-time motion control.  A slow
>parallel port could be used for the auxilliary controls.
>
>Note that the STG card is an 8-bit ISA card, and performs
>quite well, with servo update times of 1 mS or less.

Minor correction:
According to the STG manual, the DAC and ADC registers
"must be accessed as full 16-bit words, not as bytes."  The
encoder counters, each 24 bits wide, are arranged in pairs
to allow 3 16-bit reads per pair, but may also be read a byte
at a time, as I understand it.  Apparently STG is built to
accomodate a much higher update rate than we are using.

General info for the curious:
An 8-bit I/O read or write on the ISA bus requires 6 cycles
of the 8.33 Mhz bus clock, or 720 ns; a 16-bit I/O operation
requires 3 cycles, for 360 ns.  Performing 16-bit I/O via 
16-bit operations is not two, but four times as fast as 8-bit
I/O, since two of the 8-bit cycles would be required.

For a four axis system (z, y, x and a rotary table) there would be
4 24-bit encoder counters to read, and 4 16-bit (13 bit actual)
D/A registers to write, per update cycle.  Total I/O cycle time
with 16-bit transfers would be 3.6 us, using 8-bit transfers
would require 14.4 us.

None of this appears to be relevant to EMC;  with servos,
there's a lot of I/O, but it's only done every millisecond.
With steppers, I/O happens at the step rate, but one write
can step all four axis at once, assuming they are all on
the same parallel port.  I just did the math to see for myself.

Jim