Re: [CAD_CAM_EDM_DRO] Re: Pulse Gen

On Thu, 07 Dec 2000 23:17:23 -0000, Mariss wrote:

>
>Hi,
>
>I'm more than naive, I am actually ignorant when it comes to any high
>level programming language or OS.
>
>As I understand it, the problem is one of latency with WIN95/98/2K. I
>am also given to understand this latency can be as much as 20 ms.
>
>Is this a worst case time, and am I correct in assuming this would be
>a problem with any other WIN PC port, be it serial, USB etc.?
>
>The way around it may be to use a FIFO. The worst case latency time
>would set the minimum size of the FIFO.
>
>Mariss

Hi Mariss,

The worst case interrupt latency on a PC is more like infinite. That is,
some just get lost. As I said in my off list message to you: "For any PC
device driver to work reliably it must be designed to tolerate missing
interrupts." Not just late, missing. This is not just a Windows problem,
but Windows makes it worse. The underling problem is bad hardware and
BIOS design.

The following is a copy of my off list message to Mariss:

>Hi Mariss,
>
>I've gone back and read a lot of the posts on the CAD_CAM_EDM_DRO list
>on this topic that I had skipped over previously. This has resulted in
>my having a better understanding of what you're talking about, I hope.
>
>First of all, please don't spend (or let Dan Mauch spend) any real money
>getting boards made until we talk this through. I think I see serious
>problems.
>
>As I now understand it, you are proposing the following:
>
>The pulse generator circuit is the one you posted to the CAD_CAM_EDM_DRO
>files area about three months ago. The idea is to hang a string of N of
>these on a PC parallel port, where N = number of axes.
>
>The inputs of the first eight bit latch would be driven by the port data
>lines. This first rank of latches would be clocked with the strobe line
>from the port.
>
>The second rank of latches is clocked with a ~1KHz internal clock that
>would also drive the ACK line on the port to tell the PC software when
>to refill the first rank of latches. This gives us 1 ACK per 2N strobes.
>
>If I have any of the above wrong, please correct me.
>
>If this was interfaced to a conventional port on a microcontroller (like
>your Z80 example) I believe it would work. Alas, a PC (and it's printer
>port) is a very different beast.
>
>The primary problem area is the PC's interrupt system. It's design only
>makes any sense at all when viewed with an historical perspective. I
>have lived through that bit of history, close up, so I more or less
>understand it. I won't bore you with the gory details.
>
>The short version of the story is that PC interrupts are undependable.
>For any PC device driver to work reliably it must be designed to
>tolerate missing interrupts. I know that seems all wrong, but it's a
>fact of life. PC printer drivers typically do not even use the parallel
>port's interrupt.
>
>Much of what's wrong with the current state of PC hardware design is the
>result of some very bad design decisions that IBM made for the original
>PC. The interrupt system is a prime example. The only real fix for this
>is a redesign.
>
>This process is underway. You will note that the latest batch of PC
>motherboards have no ISA slots. This is a prerequisite for dumping the
>old interrupt system. The hit list also includes the printer port. The
>newer printers (and all the other stuff with parallel interfaces) all
>come with USB interfaces.
>
>I don't want to put a lot of effort into something that can't work
>reliably, and that uses an obsolescent interface. I'm not saying we
>can't make some kind of hardware pulse generator work, but I'm sure the
>hardware will need to be more complex than you expected.
>
>I'll be giving all this more thought in the next few weeks.
>
>Jeff

Jeff