[Olsr-dev] broadcast packet filtering

[Erik Tromp]

> its not bashing. - i was just expressing common insights in 
> router design that any scheme that mixes up forwarding-plane 
> with control-plane have an inherent risk of destabilizing the 
> control-plane of the system.

Your assumption is wrong, and thus all that follows - "Ex falsus
quod libet" :-) . Bmf does not mix up the forwarding plane
with the control plane. Bmf uses encapsulation, which is a modern-day
technology. Bmf does this in order to: 
1. prevent multiple copies of the same IP packet from being received
   by the application software
2. prevent the need for putting network interfaces into promiscuous
   mode, as is done in other MPR-based flooding/forwarding mechanisms.

For more information, read the README_BMF.txt file, especially
paragraph 5.

Bmf *does* perform the forwarding in user-space. In Linux-based
systems, it would be nicer (i.e. more elegant and maybe better
performing) to have the forwarding done in kernel-space. However,
current linux-kernels do not offer an implementation that filters
out the forwarding of recently seen packets. Maybe a kernel
module can be written for iptables to implement this, but that would
introduce the dependency on an external software component, which is
not my preference.

Another option is to guarantee that each packet gets forwarded to
a downstream node by at most one upstream node. This is the assumed
way of working in current Linux-kernel multicast forwarding
implemenations. This requires building of a forwarding tree, which
reduces redundancy to zero (each branch in the tree is a single point
of failure), and is not consistent with the MPR-based redundant
mechanism used in OLSR.

> not vague opinions, more experience of building and 
> supporting routers for 10+ years :-).

Same for me ;)

Thanks for the URLs!

Erik