And just reiterating: BATMAN works just fine for larger networks, just so long as you're happy to use BGP between disparate BATMAN islands. But first, ask yourself whether you really need a network with 500 routers involved in decision making on that little piece of paradise...
Cheers,
Michael
On Wed, Dec 1, 2010 at 2:52 AM, Michael van der Kolff mvanderkolff@gmail.com wrote:
They are not theoretical, just not as production hardened as BATMAN. Which is to say, finding small-time implementations of them isn't too hard - they do have working code. But their code runs on top of the Click routing system, which will never find much acceptance within the Linux kernel community due to its use of C++ (albeit with exceptions & other things stripped). In short, they have working code, but it is a proof of concept. No one has actually got a real 125,000 node system working, but the simulations of the routing protocol do make that prediction, yes.
"Real world deployments" is always a tricky and nebulous concept. BATMAN has real world deployments due to Freifunk & other initiatives, who naturally favour code that has a real chance of being accepted into the kernel. Not using C++ is good for kernel acceptance.
But yes, there have been some honours theses written around small-scale deployments with extensions to do things like hash-based lookup of names (compute the hash and make the routing protocol stop when its successor is greater than the hash value, instead of the IPv6 behaviour which is an ICMPv6 response...)
For more information, google Linyphi. Or, indeed, netsukuku. But seriously, the point is mesh networks are an open research topic, and I should imagine that many more PhDs will come out of this field - but BATMAN is very much focussed on reduction to practice - making the current ideas work as best as they can - which is why it is suitable for the networks it has been deployed into.
Cheers,
Michael
On Wed, Dec 1, 2010 at 2:18 AM, Kosta -- A Human Right kosta@ahumanright.org wrote:
Are those protocols simply theoretical? Or could they work if someone tried them?
-K
On Tue, Nov 30, 2010 at 7:10 AM, Marek Lindner lindner_marek@yahoo.de wrote:
On Tuesday 30 November 2010 15:55:16 Michael van der Kolff wrote:
It is in fact feasible to have thousands of mesh nodes - http://www.so.in.tum.de/wiki/index.php5/Scalable_Source_Routing gives a real model for something like that. Linyphi gives a real model for that - embed a MAC address into subnet address, and get routes to your immediate successor, 2nd node, 4th node, etc, and forward packets to the node closest numerically to the destination while still being under it. Of course, Linyphi is hardly production ready - more a proof of concept.
Let me rephrase then: I don't know of any mesh protocol that is being used in real world deployments and scales up to millions of nodes.
If we leave the realm of working implementations I'd like to add Netsukuku to the list of endlessly scalable mesh protocols. I quote from their FAQ:
It [Netsukuku] can be used to build a world-wide distributed, fault-tolerant, anonymous, and censorship-immune network, fully independent from the Internet. [..] The number of interconnected nodes can grow endlessly.
Cheers, Marek