[2132] in Commercialization & Privatization of the Internet
Re: Trickle Down? How About Bubble Up?
daemon@ATHENA.MIT.EDU (Dave Crocker)
Tue Jan 28 01:21:17 1992
To: "Miles R. Fidelman" <mfidelma@diamond.bbn.com>
Cc: com-priv@psi.com
In-Reply-To: Your message of Mon, 27 Jan 92 20:29:35 -0400.
Date: Mon, 27 Jan 92 22:20:06 -0800
From: Dave Crocker <dcrocker@Mordor.Stanford.EDU>
Miles, It would be unreasonable to ignore the amount of overhead that
tcp headers consume on a low-bandwidth line, especially. So I agree with
the direction of your concern, but there was an advance, a couple of
years ago, which fixed the problem without requiring moving to a
different transport stack. It is called header compression and
basically involves having the code notice exactly what you said: Most
of the header bits are the same, for consecutive packets. Hence, after
you send them once, then the sending side of the link merely needs to
signal that the relevant header bits on "this" packet are the same as
the previous one. (How's that for an oversimplification?) I think the
result is a header overhead of 1 bit, plus a fraction.
The approach that you have described basically leads into requiring
a gateway, application-level, transport-level, or the like, and
gateways are courts of last resort, to be avoided when possible.
Dave
(
