RHEL3 ES: problem with demand dialing...

I have setup my new computer, running Red Hat Enterprise Linux 3 ES, to 
do demand dialing since it serves two other machines and all must share 
the same dialup line.

It is set to start at boot, but to drop the connection every time 600 
seconds go by with no transmission. (Spammers can keep the connection up 
by sending me a spam every 590 seconds, and I see this happen sometimes).

But _the problem I am concerned about_ is that sometimes the connection 
drops and pppd does not seem to notice. I.e., users want a connection, 
the little green light in modem-lights applet is lit, and everybody 
seems happy. But if I pick up the telephone plugged into the modem, I 
get dial done. So pppd did not notice.

If I look at the /var/log/messages, what has usually happened is that 
pppd timed out and voluntarily broke the connection, though it seems to 
have failed to turn off the modem-lights light (which AFAIK is governed 
by the presense of /var/lock/LCK..ttyS4). Now in the distant past (RHL 
6.2?), when I did not run demand dialing, pppd had a bug that did not 
always clear the lock file, but a later version of pppd fixed it. I 
would like to think this bug has not returned.

I can cure this by killing pppd, removing /var/lock/LCK..ttyS4, and 
doing ifup myISP. But if I am out of town for a week, this would not be 
a good thing.

Is this likely to be a configuration error? Or a bug in pppd (or wvdial) 
or what? I do not wish to just tolerate this as it happens almost every day.

-- 
   .~.  Jean-David Beyer           Registered Linux User 85642.
   /V\                             Registered Machine   241939.
  /( )\ Shrewsbury, New Jersey     http://counter.li.org
  ^^-^^ 13:40:00 up 6 days, 9:55, 2 users, load average: 4.09, 4.02, 4.04