Comments on Analogue and Digital: Argot Meets Contiki

If you want reliability over UDP you typically end...

2009-10-07T18:07:11.795+11:00

If you want reliability over UDP you typically end up recreating most of TCP anyway (but typically in a worse way, that hasn't been thoroughly debugged in the same way that TCP has), and you lose all the interoperability benefits of TCP.

TCP has its share of problems over wireless links, but they mostly have to do with the performance of large bulk data transfers, which we are unlikely to have in low-power radio networks anyway.

The TCP header overhead can be reduced through header compression instead of through defining custom last-hop protocols over UDP.

you're correct that I'm using UDP for the ...

2009-10-07T09:34:49.105+11:00

you're correct that I'm using UDP for the RPC. This was done as a proof of concept more than anything. In the past I've normally used TCP, Pipe or even HTTP (tunneled) transports.

In the Smart Energy 2.0 and 6lowapp discussions various people have said that TCP would be too much overhead for a 6lowpan network. In those situations they suggest that UDP would be better. Do you think that's a fair assumption?

From what I understand, you are using UDP for the ...

2009-10-06T22:32:29.537+11:00

From what I understand, you are using UDP for the RPC, which means that you will need to implement your own sequence numbers, round-trip time measurements, and retransmissions. Have you thought about using TCP instead? That will give you a reliable channel with a much lower overhead than a custom-TCP-over-UDP.