🚀 go-pugleaf

I have a 486-33, at 8 MB with 1 MB SIMMs,
and probably destined to stay there:
looks like BIOS doesn't like 4 MB SIMMs.  It's an old
ASUS-TEK motherboard, 128K cache.  System has
two disks: one IDE, one SCSI.  IO board has the usual
two serial, one parallel port; no idea what its  jumpers do,
so I put the mouse on COM1 and left COM2 talking to air.
I put in a U.S. Robotics 33.6 internal modem, set its jumpers
for COM4 and IRQ 5, tested the modem with HyperTerminal,
configured the Win95 stack to talk to my ISP (pacbell),
and installed Netscape 3, News Express, and Eudora Lite.
Tested it: everything performed quite well.
Connections connect, packets flow, I can
surf the web, read news, and send mail.
Nice little Volkswagon on the information superhighway :-)

All great, except that when I'm receiving packets flat out at 33.6,
e.g. during a large file download or when fetching the news headers
from a high volume newsgroup, after a few hundred K
the packets stop flowing.  The cute simulated modem
lights on the task bar are glowing green, the download
stats are marching away, Netscape's ftp gives download
speeds at modem speeds or slightly higher, and then
suddenly the modem lights are red -- and at that point,
I can't get packets to flow from ANYTHING: ping,
web, anything.  However when I disconnect and reconnect,
everything is fine again.  And everything stays fine as
long as I don't do something that causes a lot of packets
to arrive at modem speeds.

I have found that lowering the modem speed to 14.4 (well,
setting the modem and comm port to 19.2) pretty much
makes this problem disappear.

I tried playing with the 16550 send/receive buffer settings;
doesn't seem to make any difference.

Is this a known problem with the Win95 stack?  Is there
a well known fix?  Patch from Microsoft?  Someone
else's stack?  Some parameter  to tweak?  I'll try anything.
Just don't want to be stuck at 14.4 on this box and
probably still unreliable, occasionally.

Any ideas welcomed!

Thanks,

-- Jon