IP Switches

[Bdale Garbee <bdale@col.hp.com>]

> Since you've worked with the DE so extensively, do you have any thoughts
> on whether it would be practical to daisychain DEs together to add
> ports?

Depends on what you mean by practical.  I'm of the opinion that the cross-link
between switches that are "part of a site" needs to be faster than the links
being supported for the propagation time across the switch to be acceptable to
me.  That doesn't mean it won't work, and in fact the G8BPQ code supposedly
supports the NRS-style interface on the "console" port of the DE to allow the
DE to be matrixed in with other DE's and/or TNC-2's with NET/Wrong in them.

> I realize it's a kludge, and the speed between the boxes would
> limit cross-DE links (can the serial port be driven faster than 19.2?),

I don't believe so.  The internal serial port on the V40 is very much like an
Intel 8251, and while I don't have my data books handy here at work, but I've
never run the port faster than 9600, and doubt it will go much more than 19200.

> but it would be a way to build an expandable switch with minimum up-front
> cost.  (And, with street prices for DEs being in the $279 area, a pair
> of them are still cheaper than the Grace card)

But what you end up with still isn't a Grace card... the potential for going
to ever-faster data rates in the future, lots of RAM space (512k by default,
2Meg with 1Mx4 ZIP DRAMs) for routing protocol tables and new features (or
bootloading the executable image ala satellite operation with new ROMs), and
no degradation of throughput because the port you need to go to is "in the
other switch"... plus the appearance of a single node, not multiple nodes, in
the NET/Wrong routing tables, etc...

To the same extent that you can build multiple port nodes with TNC-2's, you can
build them with DE's, or PC's, or anything else you care to imagine.  It's less
clear that you ( or I ) will be satisfied with the result.

Bdale