NetScape standards, 2.0 features, and purple pages

[Michael Gersten]

Holger writes:
><soapbox>
>
>Yes and no - this is really two-fold.
>Of course braind-dead TCP/IP isn't very useful on either really-
>high-speed or low-bandwith networks, reliable or not. See the
>ATM standards commitees and vendors bending over backwards in
>order to achieve high throughput over these nice 155/622 mbit/sec
>links..
>
></soapbox>
>	Holger
>  ___ 
> /\__\  Holger Hoffstaette              @work: hhoff at cube.de
> \/__/  Cube Informationssysteme GmbH   @home: hhoff at flop.lb.bawue.de

Ok, what is it that makes TCP/IP unusuable on high speed, or low
bandwidth lines?

For low bandwidth, remember that header compression normally cuts the
overhead to 3 bytes per packet, and that Jacobson has gotten it down
to an average of just over one bit per packet (when he could do his
own error correction, not relying on TCP's error correction).

For high speed lines, aside from the 16 bit, 64K window size, which
could trivially be made a 32 bit window size, what else needs to be
done?

The only problems I know of are the routing/addressing problems, and
people are working on those.

		Michael