Re: [homenet] Updates to Homenet Architecture Principles doc

[Juliusz Chroboczek <[e-mail]>]

> So even though link-local multicast may be part of the IPv6 base spec,
> it may be desirable to avoid use of multicast traffic where
> possible. e.g. a routing protocol could perform initial neighbor
> discovery using multicast, but then switch to unicast when maintaining
> individual neighbor associations on the longer term, or for exchanging
> information with specific neighbors.

Ah, I see.  Yes, that makes sense.

>> Then I think I don't understand what is meant by "multi-path support".  Is
>> that merely the ability to switch to a better route if one becomes
>> available?  Or is there something more to it?  Could you please clarify?

> "The inclusion of physical layer characteristics including bandwidth,
> loss, and latency in path computation should be considered for
> optimising communication in the homenet."

> I interpret the above text as requesting an ability to perform load
> balancing over equal cost paths, potentially taking into account packet
> loss and link quality when selecting which stream to send over which link.

I don't.  I interpret that as meaning that the routing protocol should be
able to pick a path using more refined criteria than just hop count.
E.g. to prefer a 1Gb/s link to a 100Mb/s one, or to prefer two wired hops
to a single wireless one.  Doing that is not too difficult, and yields
dramatic performance improvements in some fairly simple, realistic
topologies.

Load sharing is a completely different headache.  Done carelessly, load
sharing causes issues with increased latency and jitter, and causes
intermittent connectivity problems that are very difficult to troubleshoot
if one of the links is unreliable.  It also has ghastly performance if you
balance across wireless links without taking radio interference into account.

While it's certainly an area worth experimenting with, I'd be cautious
about requiring load sharing in Homenet unless somebody knows how to do it
right in the general unadministered case.

-- Juliusz