[Babel-users] Fun with metrics [was: Babel questions]

[Will Glynn]

On Oct 20, 2010, at 10:32 AM, Juliusz Chroboczek wrote:

>> Well, I'm intending to use 802.11, but in infrastructure mode
>> (AP/station) rather than ad-hoc.
> 
> I'm wondering if you're not over-complicating things.  Are the STAs
> actually forwarding packets?  If they aren't, then why don't you just do
> static routing on the STAs, and run Babel on the APs only?

Yes. Some locations have line-of-sight to reach multiple already-connected locations, and the cost of adding extra radio gear really isn't that high. I want non-tower devices to participate in routing.

>> Packet loss happens, and with TCP/IP traffic, should be avoided -- but
>> like city driving, sometimes the best route is congested and of
>> questionable quality, and sometimes it's the only route to your
>> destination.
> 
> Hmm... ETX assumes that the network is not congested, and that the only
> cause of packet loss is due to the phy layer.  Are you trying to design
> a routing metric that takes both congestion and phy-layer issues into
> account simultaneously?

I was intending to refer to physical vehicle traffic in my analogy. Such a metric for network traffic would be really cool, but I have no hopes of finding one.

>> Good point. I think I also need it to be aware of traffic levels --
> 
> Heh.  Google for ``load-aware metrics''.  Careful about failure of
> isotonicity.
> 
>> just a general feeling of if I should try.
> 
> You definitely should try.  Playing with metrics in this way certainly
> counts as interesting, original research, and it's stuff that we can
> only do because we know that Babel is robust in the presence of weird
> metrics (i.e. the proof of loop-freedom of Babel doesn't depend on the
> exact metric being used).

That's one of the things I love about Babel. The specification makes it clear that I can do crazy stuff without breaking things.

I'm planning to keep the metrics all additive, just twiddling the calculated cost. Running in infrastructure mode and enabling layer-2 isolation pretty much reduces the wireless portions of the topology to a series of point-to-point links, which simplifies things considerably. The wired portions (e.g. the ethernet interfaces that glue together different radios at a location) shouldn't need any tweaking.

> On a related note, please check babelz.

Will do!

--Will Glynn