jim@baroque.stanford.edu (James Helman) (08/15/90)
pbiron@weber.ucsd.edu (Paul Biron) writes:
I got one reply from someone at Telebit, who explained
at little bit about PEP itself, and specifically about
Telebit's and SLIP. The jist of his message was that
Telebit's don't handle SLIP very well -- it causes
thrashing between small and large packets
Telebit (or someone they've
donated a mass of TB's to) is working on ways to
improve their performance on SLIP
This is probably Van Jacobson's CSLIP (SLIP + header compression), the
beta version of which has been available for more than half a year (from
ftp.ee.lbl.gov). I believe the header compression is supposed to make
single character TCP/IP packets much smaller, hopefully small enough to
fit into one of Telebit's micro-short packets, rather than requiring a
largely unused 256 byte long packet.
I have CSLIP running on our Sun-3's (SunOS 4.0.3 w/Rayan
Zachariassen's driver for streams). Now, I'm looking for modems.
Questions:
1) Does anyone have any experience running the beta CSLIP on Telebits? If
so, how well does it work (throughput and interactivity) and with what
models, settings, etc?
2) Will SLIP's successor, PPP, have different modem requirements?
3) Does CSLIP beta work with SunOS 4.1?
4) The System V streams stuff in SunOS 4.0 slowed down the serial
ports considerably. How much CPU horsepower does it take to service a
port around 9600 or 19.2K? Is SunOS 4.1 any better?
Thanks,
Jim Helman
Department of Applied Physics Durand 012
Stanford University FAX: (415) 725-3377
(jim@KAOS.stanford.edu) Voice: (415) 723-9127henry@zoo.toronto.edu (08/17/90)
> I believe the [Van Jacobson] header compression is supposed to make >single character TCP/IP packets much smaller, hopefully small enough to >fit into one of Telebit's micro-short packets... Yes; in the most common case, he gets a TCP/IP header down to 3-4 bytes, which is not bad when you consider that two of those are the TCP checksum (which is non-removable and non-compressible if you really want end-to-end checking, and yes, you really want end-to-end checking). So a one-character packet will fit. The Usenix terminal room in San Diego, I think it was, used PEP and an alpha version of CSLIP begged from Van for its Internet link. It was not wonderful, but it was tolerable. You should not expect beautiful TCP/IP behavior from a PEP modem; TCP/IP really wants a really truly full-duplex channel. The next Usenix terminal room, Baltimore, used V.32 over a T2500 instead, which was considerably better. >2) Will SLIP's successor, PPP, have different modem requirements? I believe Van had considerable input into PPP, so it should not be worse. (I haven't studied PPP carefully.) I doubt that it will be better, except that it will probably get better manufacturer support since it is a real standard and (in most ways) a superior one.