Time to kill layer 2

[Phillip Hallam-Baker <[e-mail]>]

This morning I spent an hour debugging the network to print out two
class projects that were due. Some points:

1) My ability to debug the network is better than 99% of the population
2) The interaction of Bonjour, DHCP and auto power saving is unfortunate
3) Things should still work after I have been away for a week
4) If vendors want to be selling all that IoT gear, they have to solve
these issues.

5) I want someone to blame. Right now when the network doesn't work, I
don't know who is the cause. I want one point of contact. Whoever is
that point of contact will get most of my networking money.


One of the biggest headaches in debugging is that 'smart hubs' are
not. They are actually very stupid. They make assumptions of network
topology that are not true. Another is the unfortunate implementation
of DHCP.

I don't use SNMP for a simple reason - it is not available to most
ordinary people. I want to understand networking for the 99%, not the
IETF 1%-ers.

All this networking gear is presented to me as black boxes over which
I have absolutely no control (which is fine-ish) and no visibility.

What we have today is the product of a historical process. I remember
the days when Ethernet ran on 10BaseT. But I installed my first switch
30 years ago and it has been a switched protocol for 20 years now.


It seems to me that there is a business opportunity for any vendor who
takes the rather obvious step of simplifying the system.

People talk about 'IP everywhere' and 'IP end-to-end' which is rather
odd when you think about the fact that virtually every local network
uses MAC addresses for routing.

One of the reasons that IP won against OSI was that it was simpler.
Applications ran on top of the IP layer with only TCP inbetween. Of
course these days we do have a Presentation layer, Web Services run on
HTTP. But unlike the OSI presentation layer, ours does not introduce
extra moving parts.

It seems to me that if we really believed in IP everywhere and IP
end-to-end we would insist that network switches be IP routers that
can be managed using BGP/OSPF or at least routing tables rather than
heuristic devices that try to guess where packets should go based on
goat entrails, phases of the moon or whatever they use.


What should have happened many moons ago was that DHCP should have
become a bidirectional protocol or a bootstrap to a bidirectional
protocol. So when a printer joins the network, it authenticates and
tells the network what it is. And this is all defined in one set of
specifications from one organization, none of which assumes that
security is an 'advanced', 'optional' or 'enterprise' feature.

Instead we have an ad-hoc layer trying to achieve the same result in
peer-to-peer fashion. A similar approach works for frogs as a
reproductive mechanism but only at the species level. It certainly
does not work for the individual ova which may or may not connect to
the printer it is trying to use to print the kids damned homework.


Seriously, the fact that things have scaled thus far and the 1% can
get them to work does not mean that we can get to the next level
without a serious rethink of the local network architecture.

The type of device I think we need would be first and foremost an IP
router. It would have ethernet plugs on the box and use ethernet layer
1 specs. But when a another 'True-IP' device was plugged in, it would
quickly negotiate a direct IP connection, oh and with proper 64KB
packets. It would also, authenticate, announce and turn on link layer
encryption.

Such a device would also be a legacy router. It would fake all the
signals necessary for a legacy ethernet device to function. It would
also be responsible for maintaining the local information for the
network service database and intercommunicating with other hubs to
achieve a global network view.


The net result of all this would be that I would never ever need to
install another printer (no, it is not actually necessary for every
stupid printer to have its own stupid printer driver). Opening the
'printers' folder would automatically show every printer that is on
the network or can be woken from slumber by the hub it connects to.