To IP or not to IP (to the end nodes)?

The discussion about pros and cons of running IP protocols to the end nodes in capillary mesh networks is as hot as ever. And as ever there are two camps: the IP camp absolutely wants to do it pointing the benefits and the no-IP camp that says "do it when needed, but not as the default option".

Speaking of benefits - there is effectively just one: no application gateway, or in other words, no translation of the application layer when bridging a low power capillary mesh to a LAN network or the Internet. If an end node on a low power mesh network runs IP protocol, the application may use it and the gateway that connects this low power network to a backbone will just be a simplified bridge (or, more precisely, a border router).

On the other hand when the end node does not run IP, but some other transport protocol, the gateway connecting to the backbone must understand that protocol and do some sort of translation. So the gateway becomes more complex.

The reason for not running IP on capillary mesh networks is capacity. Non-IP transports are more efficient. Packets are more compact and the network can accommodate more of them. That translates directly into how many nodes (or, more precisely, how many messages per second the network is able to carry). More compact packets take less time on air, so more of them can fit on a single frequency at a given time period. Shorter packets also mean it takes less energy to generate and transmit them, so, for example, the transmitter may never need any permanent power source or a battery. Energy harvesting transmitters work like magic: a switch on a wall harvests kinetic energy when pressed. A sensor on a wall has a small photovoltaic panel sufficient to power it forever. An asset tag harvests radio noise and transmits an encrypted ID periodically.They are all magical products: wireless, no power required, work forever and do work reliably. Nothing like that is available today running an IP stack. IP requires more energy.

So the argument is between complexity of a gateway and performance of the network and availability of essential features like energy harvesting. From the user perspective the choice will always be clear: the market will always be voting for features and capacity. Gateway complexity? Well, vendors will take care of it eventually. After all that complexity is not that big. 

Also the IP to the end node proponents are very often short of arguments. I heard opinions like "this is the Internet of Things so it must run IP" or "everything will be IP" etc. The truth is these arguments are empty, they do not mean anything from functionality or product perspective.

Can aircraft carry shipping containers? Sure they can. Are we using aircraft to carry shipping containers? No, because it is not efficient. Does it take extra effort to pack cargo into anther type of container that aircraft can carry more efficiently? Yes it does. It is called ULD, or Unit Load Device. Overall it all makes sense, from the capacity and power budget perspective. And some smaller aircraft do not use any containers, as they are simply too small to allow for such packaging. See the analogy?

Wireless is different. It is different because of the physics that define behaviors of radio transmissions. Applying wired way of thinking to wireless is just wrong. You cannot attach wings to a train and expect it to fly...