Re: A little more information

[Tom Droege <tdroege@veriomail.com>]

Arne and all,

Let me respond point by point.

1)  I am sick of working on 1).  But I will have a another go at it in a 
day or two.  This requires writing some test software so that I can put the 
board in a loop and put a scope on it.  There is always the chance that I 
will find a simple problem.  But I am not optimistic.  I have struggled 
with the problem too long.  The memory board is bigger and more complicated 
than it needs to be.  Always a recipe for disaster.

2)  Unless someone knows of a byte parallel PCI card with 16MBytes of 
memory this is a big project.  Bigger than other solutions I have in hand.

3)  As I understand it, the ECP contains a FIFO.  Possibly someone on the 
list had interfaced to the ECP and understands it.  If so please speak 
up.  At the moment this looks like the simplest solution, but it does need 
software.  As Chris points out there is standard software for this.  Note 
that there are high speed scanners that work into the parallel port, so it 
must be possible to go at our rates.

4), 5)  We have this type of solution available.  Bill Haynes has designed 
a serial card that will go at 2.5 GHz.  This over a short Base 10 
cable.  There are cheaper drivers that will go slower and should be able to 
run over longer distances.  This requires that we build a board that plugs 
into the present DB-25 connector and has a small board where we put some 
sort of control chip which then drives the serial cable.  Bill has done 
similar things before so it should not be much of a problem.  We are all 
lined up to start this project.  Then Chris suggested that the parallel 
port might work.  The PCI card that goes into the computer exists and has a 
FIFO in it.  No problem going at our 3.2 MBits since it is working at 2.5 
GBits.  ThePC end card is already designed with a dual driver.  It can 
either drive a Base 10 cable, or you can solder in a Laser driver and go 
over fiber optic.  Then there is no practical distance limit.  I consider 
this the delux solution.  But I am pretty confident that it will work.

I very much agree about retrofitting all Mark IVs to run with the same 
hardware system.

My vote is to have a college try at 3) and if that cannot be made to work 
easily to go to the PCI serial solution.

Any one with a Mark IV or a Mark IV test set up want to have a go at 
2?   Note that if someone other than Chris (or one of the Mark IV owners) 
wants to have a go at 2) we could arrange for Chris to send the test set up.

If anyone is up for it, I will build a kludge that plugs into the Mark IV 
driver end and puts the signals on the right pins to drive the parallel 
port.  I consider this mostly a software project.  One says preps the 
parallel port to accept data, says "start scan" to the Mark IV, and then 
looks to see what his software did with the 16 MBytes that the Mark IV sent.

The other way to do this is for me to build the kludge here, set up a 
computer, and have someone send test software that will prove that it 
works.  This is less desirable for me since it interrupts the Mark IV 
production.  There are more of you that would like to have a Mark IV.

What do the rest of you think?

Tom Droege


At 07:13 PM 8/14/01 -0700, you wrote:
>The important point is that the average datarate is 400KB/s,
>within the range of standard EPP/ECP port protocols (though
>outside the range of the standard parallel port).  If you really
>want to get rid of the memory card, I see several directions
>you could head:
>   (1) figure out the race condition and repair the current card.
>       This is probably the first thing to try; all other solutions
>       involve considerable hardware/software.
>   (2) replace the ISA card with a PCI version.  This still has
>       the long cable problem.  While Tom made our cablelength
>       work, it is still a construction detail that should be
>       avoided.  (Though I'd love to see someone design such
>       a board, as there are other groups that could use a basic
>       parallel port interface onto the PCI bus.)
>   (3) Drop all cards and go directly into an ECP/EPP parallel port
>       DB25 interface.  Probably fairly simple from the hardware
>       point of view, though a FIFO would almost definitely be needed.
>       Some software development.  Still long cable.
>   (4) Put an ethernet interface on a local interface (like the Stamp board)
>       and do all communication over that.  Fugitsu makes such chips,
>       fairly easy to work with at the 10BaseT level.  However, major
>       hardware and software on the Mark IV end.
>   (5) Put an embedded controller near the Mark IV, use something like
>       the parallel port (I'd think of a standard 8bit I/O port on
>       the controller rather than a IBM PC-style printer port), then figure
>       out some method of shipping the data back to the host computer.
>       Again, ethernet is the most likely, though I don't know of any
>       cheap controller card with on-board ethernet.  This is probably
>       the most practical path, and easiest to farm out to some list member
>       who is interested in low-level programming.  The advantage here is
>       that most embedded controller boards come with built-in libraries
>       of functions, especially if ethernet is included, so the software
>       effort decreases.  Note that a standard ISA/PCI bus computer could
>       be used, but the 8bit I/O is easier done with a standalone card IMHO.
>Tom loses control of some aspects of the project unless path 1 or 2 is 
>selected.
>Whatever approach is used, I highly, highly recommend that all systems are
>upgraded to the same solution as soon as is practical after the first such
>system is shown to work.
>Arne