I don't know whether to argue this on the grounds that N^2 grows faster than N
and that the memory and processing capabilities of any single routing node are
insignificant next to the power of flap propagation
It's more fun to suggest that not only should the Internet work in theory,
it should work in practice, and in such a way that it can continue growing
along several axes (number of users/links/nodes, time-online per user,
amount of traffic moved per user per unit of time, ...). This suggests
that cheap technology SHOULD be used. Unfortunately, the people who argue
for "home appliances" have fallen victim to the theology of capex as the
one true God, and should invest in some dotcoms/dotnets that work along
those lines, while they still can.
The reality is that upgrade and replacement costs in an expanding
corner of the Internet tend to be large, and downtime represents
a substantial real operational expense in the presence of modern
SLAs (not to mention opportunity costs). Understanding those costs
is as important as choosing equipment that works well enough today.
This makes even home appliances appear expensive to own compared with the
physical goo inside them, and there is no magic trick around high TCO in
a busy part of the Internet.
Forgive me for discussing engineering principles on NANOG. I know
they're off-topic, and will shut up now.
PS also sorry for the trendy acronym meaning total cost of ownership