Re: [core] Fwd: I-D Action:draft-eggert-core-congestion-control-00.txt

[Carsten Bormann <[e-mail]>]

On Jun 24, 2010, at 10:22, Lars Eggert wrote:

> http://www.ietf.org/internet-drafts/draft-eggert-core-congestion-control-00.txt 

Hi Lars,

this is very useful early input (and might help us prevent late surprises at the IESG stage).  Thanks!

One thing we need to be vary careful about is the burden we place on implementers/implementations.
After all, these are constrained nodes, and both code and table (data) sizes are constrained.

What could happen is that we mandate something wonderful that then nobody implements.

Keeping RTT information about other nodes creates a table that other protocols such as ND or RPL take great pains to avoid.
There should always be a mode of operation (not necessarily optimal) that avoids creating tables such as this.
More power [sic!] to nodes that can implement them, they will have better performance.

I do like the credit proposal.
If we could phrase this in a way that helps a very constrained node to keep tabs on its battery usage as well, this might be a winner.

Most importantly though, we need a lot of time to answer these questions well (there is a reason transport development takes time).
If we can find minimal requirements that can be put into the document and leave open the way to plug in better algorithms, we can have fruitful evolution.

With some danger of creating a controversial strawman for this: We could say "at a minimum use a fixed initial RTO of 3 s and exponential backoff and don't run more than a couple parallel requests, or keep RTT and credit information and do something better".

Gruesse, Carsten