🚀 go-pugleaf

Okay, simple question!  Linux v2.0.33 & pppd 2.2.0f with a external
modem using hardware handshaking, and 115K2 data rate...

How would I know if I am having serial overruns (i.e. not guessing
from transfer speeds or pauses but actually *know*)?  

I'm trying to get our system flowing nicely under ppp, but I can't
really tell whether or not I'm loosing out due to overruns.

The not so simple bit!  The background information; the reason I
suspect overruns is as follows:

I seem to have some very strange problems w.r.t. overruns using
Linux, X2 48K connection & the USR BBS with Minicom -- with very fast
transfers (i.e. very highly compressable data) I start getting,
whenever the CPS is around 9000 or above and when the system writes
the data to the hard-drive, overruns...

...what is strange about this is that it's...

1. hdparm & irqtune are set up accordingly.

2. for less than 9000cps, RX data xfer is *rock solid* -- I can do
virtually anything with the computer and nothing nasty happens, no
overruns etc.  I can log onto X, start up netscape, frantically swap
between text console and X, etc. etc. etc. and *no* overrun problems.

3. *other* hard-drive accesses doesn't cause problem, the problems pretty
much *only* occur when the *downloaded* data is being written to the
drive (X killing the system with memory swaps doesn't seem to hurt --
writing 9Kb/ps of the downloaded data on the otherhand *does*!)

4. changing the RX FIFO IRQ trigger level makes *no* difference (and
this is a 16550A compat. UART)


Hardware handshaking *is* definitely enabled.

Having looked at serial.c etc. in the Linux kernel, I'm begining to
get very suspicious.  It looks to me like the design of serial
handling under Linux doesn't really give the handling of hardware
handshaking a particularly high priority.  With the problems I
am having it looks as if any idea with respect to perhaps lowering
RTS (?) would always come much too late, i.e. the latency caused to
the tasks that should be handling hardware handshaking induced by the
respective disk writes is just too much.  It seems that RTS could
really do with being played with at a low/IRQ level, instead at such
a high level?  Am I talking complete rubbish here?

Upgrading our hardware is not an option sadly (not unless there's a
fantastic solution that involves a drop-in perfectly
hardware/pin-compatible replacement for a 16550A chip that'll solve
all the problems and doesn't cost an arm & a leg!).

Anyone any thoughts/comments/ideas on this + also, of course, how to
detect if serial overruns occur with ppp?

Ta-ra,
  Julie.

-- 
*** IMPORTANT: If replying, leave e-mail address as is, it's correct! ***
"Here's a pin and here's a pingle, porcupines are always single,
 poor ol' 'pines they mustn't mingle, pingle pangle pingle."
[singing] Marvellous Mouse Organ Mice (Bagpuss: 1970s UK Children's TV)