Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!hplabs!pyramid!csg
From: csg@able (Carl S. Gutekunst)
Newsgroups: comp.protocols.misc
Subject: Re: SLIP: Why are people so reluctant to use it?
Message-ID: <130675@pyramid.pyramid.com>
Date: 17 Oct 90 16:10:34 GMT
Sender: daemon@pyramid.pyramid.com
Reply-To: csg@able.pyramid.com (Carl S. Gutekunst)
Distribution: na
Organization: Pyramid Technology Corp., Mountain View, CA
Lines: 32

SLIP is simple to use and quite reliable over serial links where the error
rate is low, e.g., over a TrailBlazer or MNP modem, or just plain point-to-
point RS-232. SLIP itself has no error correction at all; it just does framing
of IP packets, which *do* have a checksum. There are timing problems in pre-
4.3BSD-tahoe kernels that limit throughput, particular in interactive sessions
(like telnet or an X server). FTP, NFS, SMTP, and NNTP all work quite well.

It is extremely easy to use over a direct wire or leased line. It is not very
difficult over a dialup, *if* you have a modem that can autodial a previously
stored number when DTR is raised. Someone at NASA Ames (Steve Schoch?) did a
little package called slipsh (the SLIP Shell) that makes the dialin procedure
more complex, but provides password (i.e., login) security.

PPP is really just HDLC asynchronous. A proper implementation is complex, and
can be a real CPU hog. Were I to write from scratch, I'd need about 6 weeks,
and I'm an HDLC expert. Present PD versions are little more than toys; I don't
know who/when someone is going to write a better version that they are willing
to give away. (Pyramid's version, if done at all, would be based on priority
HDLC code that we already have.)

I do not see PPP is a viable replacement for SLIP. It is just way too complex
for the job. It does offer vastly increased realiability over noisy links. But
I would not consider using a TCP/IP-type protocol over those kinds of links,
anyway. If I really want to go big time, I'll run IP right over real HDLC, on
synchronous modems, the way Cisco does. Then you have something that you can
use over the new cheap 56K DDS lines. Alas, synchronous hardware on UNIX is
rare, and far more expensive than asynch.... 

(But what do I know? I thought implementing uucp 'g' protocol on the Trail-
Blazer wouldn't fly, either. :-))

<csg>