Re: flexfax: Wedged modems and reticent faxq

["Jay R. Ashworth" <jra@baylink.com>]

On Mon, Aug 09, 1999 at 04:23:46PM -0400, Dave Bloom wrote:
> We continue to have problems with our Hylafax 4.0pl2 solaris 2.6
> servers, which pump out thousands of pages of research reports to
> our clients each day:

Wow.  :-)

> 	o Our modems (we use MultiTechs ranging from the MT2834MR6
> 	  to some older models) work as expected, but over time
> 	  they 'wedge' one by one.  Only power cycling the modem
> 	  returns them to a responsive state.  I have initialized
> 	  each modem with the following commands, which should
> 	  reset the modem to a known state when DTR goes low:
> 
> 		ATE1V1Q0X4&D3&SF1&S0&C0&F9&W0
> 
> 	  This makes wedging more infrequent, but does not fully
> 	  solve the problem.

What sort of ports are these modems hooked up to, what kernel and
driver versions are you running, and if those drivers are modular and
granular, have you tried unloading and reloading one of them (ie: one
board) to see if that helps?  This might well be a serial driver issue.

It's not the easiest thing to substitution test, but if you're on
multi-user cards, hanging one modem off a motherboard port might be
useful.

> 	o The 'wedge' command does not fix the above.  (Probably
> 	  because the modem becomes completely unresponsive.)  The
> 	  following error message is repeated in /var/adm/messages:
> 
> 		MODEM /dev/sts/ttyD0b appears to be wedged
> 
> 	  Isn't faxq supposed to take the modem out of service???

There's a retry count and time, I believe.

> 	o Possibly related to the above, 'faxquit' only kills 'faxq'
> 	  if there are no currently running 'faxsend' processes.  If
> 	  there ARE faxsend processes running, faxq just keeps on
> 	  spawning new faxsends instead of waiting for the running
> 	  processes to finish and then exiting.  'faxstate' doesn't
> 	  reliably allow me to mark a modem out of service.  It
> 	  seems the whole business of modem status doesn't seem to
> 	  be working correctly.

I've seen this mentioned on the list previously; I don't think it was
you, and I don't recall seeing a good answer.

> 	  In spite of the fact that we use HylaFAX as an outbound
> 	  fax server only, I have tried running the system using
> 	  'faxgettys' and 'faxmodem'.  Neither approach solves these
> 	  problems.

Hmmm... faxgetty is usually recommended even for outgoing only machines
because it provides positive control over the modem... at least in
theory.

> We have been using HylaFAX on five Ultra 2s with 96 modems on each
> server.  I have come to the conclusion that the software is not
> production ready for the quantities of faxes that we generate.  We
> may have to abandon Hylafax andgo back to a commercial product if we
> can't get Hyla to work reliably.

Oh.  Sparcs.  Got it.  Holy Jesus, that's a lot of modems...

I'm not sure I'd use that sort of modem for that level of traffic; fax
cards might be a better bet.  This would require that someone write
drivers for such cards, but that's no longer impossible, and it sounds
like you might be able to afford to pay the right person (I'm not him :-)
to create such drivers.  Natural Microsystems has recently started to
make information available that could be used in such a project, see
http://www.opentelecom.org.

I'm not one to tell you how to spend your money, but if you did pay
someone to write tty/modem drivers for someone's fax card and GPL'd
them, I expect you'd be well thought of...

> Any ideas????  Has anyone seen these problems before?

As noted.  There's a 4.1beta out; if you have the slack, it might be
worth investigating whether that solves some of your troubles.

Pulling stty info on the ports before and after they wedge might be
useful too, as might checking where the processes are hanging, if at
all, with top(1) (which can tell you which kernel routine a process is
waiting on).

I'm not a pro (unlike, say, Darren, or Nico :-), but I hope this
helped, at least a little.

Cheers,
-- jra
-- 
Jay R. Ashworth                                                jra@baylink.com
Member of the Technical Staff     Buy copies of The New Hackers Dictionary.
The Suncoast Freenet            Give them to all your friends.
Tampa Bay, Florida     http://www.ccil.org/jargon/             +1 813 790 7592