>From: Jan Morales <morales@cs.UMD.EDU>
>David Gilbert wrote:
>> 	Having got a pc with 16550's for the express purpose of
>> driving serial ports at high speed, I am a tad dismayed that I'm
>> getting silo overflows on the system.

>I had the same problem running 16550s on a 486/33 with an IDE hard
>drive.  Then I got a SCSI hard drive and a 1542 ISA SCSI controller
>and the silo overflows disappeared.  (And I'm running a 28.8 modem
>57600 modem-to-host.)
>
>I usually got the silo overflows while downloading, so I figured
>having the serial port and the IDE disk busy at the same time caused
>the problem.  I know SCSI is not an inexpensive solution, though...

If this is the case, I do not understand why.  The 16550 (>= "A" assumed)
has _hardware_ flow control!  If the chip's input buffer should fill
while the CPU is otherwise occupied, it should simply stop the sender
with the agreed upon hardware hand-shaking and notify the system via
interrupt to drain the buffer.  Can the driver not tell how much silo
is left?  If it can, why does it not just leave the sender stopped and
the buffer uncopied until there is enough silo?

Max