John Romkey (romkey@BORAX.LCS.MIT.EDU)
Fri, 18 Apr 86 23:22:41 est
Date: 18 Apr 1986 18:19-EST
Accepting IP on an unprotected (ie no link level) line could
be pretty messy. I would suggest that a reliable async link level
would be appropriate which does error detection and retransmission
if you go down this path.
Actually, I've used SLIP, which doesn't do any error detection that
the serial line doesn't do, over a 1200 baud dialup several times
recently, between a PC at home and a VAX ant MIT. Error rates aren't
bad at all. There are some bigger problems, though:
- timeouts in TCP/IP implementations. Many programs seem to
expect fairly high speed connections. They just don't hack
1200 baud very well. Retransmitting a 500 byte packet over a
1200 baud line is a disaster! Things are better now than
they were two years ago when we did some experiments at MIT
with a (really stupid) serial line protocol and some dialups
on a gateway (which Rob mentioned a couple of messages
back), but most TCP implementations just aren't up to
performing well over networks whose speeds range from 1200
bps to 10Mbps.
- addressing. This is an issue SLIP doesn't, um, address.
Suppose you've got a bunch of dialups on a gateway, in a
hunt group. The IP address of each dialup pretty much needs
to be set by the gateway, not the machine that's dialing in.
SLIP requires that both ends have "preordained" addresses
and already know them, but in this case, the machine that's
dialing in has to figure out its address. I have a few ideas
for fixing up this problem but I'm not sure how practical it
is to implement them.
(BTW, I'm only picking on SLIP 'cause it's there and people
are implementing it...)
Also, you're going to turn the TAC into a real gateway, which
might be more than it can handle. I don't really know.
I don't mean to be discouraging, I think it's a great idea,
being able to send raw IP datagrams through a TAC, but there are some
problems to think about. If someone were to implement SLIP for the
TAC, I've got IBM PC software they could use right now. If someone
does some other serial protocol, I'll implement it, too, provided that
it's not too monstrous.
Is there anyone out there working on this sort of thing right now?
This archive was generated by hypermail 2.0b3 on Thu Mar 09 2000 - 14:36:06 GMT