Building the Technology Stack for Internet Freedom (Gigaom)

Posted Feb 20, 2011 17:25 UTC (Sun) by sturmflut (subscriber, #38256)
Parent article: Building the Technology Stack for Internet Freedom (Gigaom)

The problem is not how to build a mesh out of arbitrary nodes. There are hundreds of solutions for that, at least two (IEEE 802.11 and batman-adv) are even part of the standard linux kernel. The major problem are scalability and higher-level-services, like address assignment, naming etc.

To my knowledge to this date there is no single solution which scales up to thousands of nodes and can cover a larger area. At some point all those networks do no longer scale and need to be backed up by a larger network running on wires - like the Internet. Mesh networks built on top of standard WiFi hardware also suffer from bad performance, since passing the same frame from one node to another blocks the radio channel multiple times - for all nodes. See the "Hidden station" and "Exposed station" problems.

If wireless meshes were the solution, anybody would already use them. The technology has been there since the eighties. In reality there are only a few installations for some special cases and the rest runs on wires.

Building the Technology Stack for Internet Freedom (Gigaom)

Posted Feb 20, 2011 18:47 UTC (Sun) by wahern (subscriber, #37304) [Link] (1 responses)

You would think that with things like MIMO and beamforming you could mimic wired, point-to-point communications over the air, especially for fixed nodes.

AFAIU, the theoretical limit to mesh networking is that the mean throughput of a mesh network asymptotically approaches zero as participation grows, because of the overhead of relaying peer data. Which seems to imply that whatever technology is being used, at some relative size it's more beneficial to fragment the mesh network. But this would be a nice problem to have, if we should be so lucky.

Building the Technology Stack for Internet Freedom (Gigaom)

Posted Feb 21, 2011 0:37 UTC (Mon) by wahern (subscriber, #37304) [Link]

Maybe it's the marginal bandwidth which each peer adds to a [perfect] mesh network which approaches zero. I need to find the paper.

Building the Technology Stack for Internet Freedom (Gigaom)

Posted Feb 20, 2011 23:47 UTC (Sun) by tpo (subscriber, #25713) [Link] (3 responses)

I already replied in the other thread here http://lwn.net/Articles/428131/ : olsr claims to have a 2000 node installation in Athens - I'm not sure whether that counts as "scales up to thousands of nodes and can cover a larger area".

Building the Technology Stack for Internet Freedom (Gigaom)

Posted Feb 21, 2011 9:47 UTC (Mon) by robert_s (subscriber, #42402) [Link]

As exciting as I find the idea of mesh networks, I do worry how well they can cope with malicious/malfunctioning users. I'm not just talking about sniffing traffic - that can be mitigated by the user, but surely a node advertising various non-existent routes can cause quite a bit of chaos.

Building the Technology Stack for Internet Freedom (Gigaom)

Posted Feb 22, 2011 9:17 UTC (Tue) by sturmflut (subscriber, #38256) [Link] (1 responses)

The problem is that there is a huge difference between an infrastructure mesh network provided by initiatives like Freifunk or Athens Wireless, and a true mobile ad-hoc network like Commotion wants to be. At the end they are both summarized under the term "mesh".

Athens Wireless and Freifunk build infrastructure using fixed nodes and directed radio links, and connect clients to it. The clients are not part of the mesh itself, so Athens wireless may have ~5000 nodes in total, but only 1000 of those are mesh nodes. The whole structure is mostly fixed, with a central organization handing out addresses etc. At the end of the day those infrastructure mesh networks are not different from wired networks or cellular networks. There is hardly any difference between a directed radio link and a wire.

A true ad-hoc network spanning a whole country or even a whole continent, built out of highly mobile nodes without any central administration, is a completely different type of problem. It does not compare to any of those existing "mesh" efforts. The overhead is immense, and you could certainly not cover an area larger than a couple of square kilometres without structured routing and wires.

Just imagine a data packet traveling from New York to Google in Mountain View and back, a thing happening trillions of times every day. The total distance is about 5800 miles. If that distance is to be covered by mesh nodes with WiFi interfaces, and you manage to max out the WiFi Range at 300 metres between two nodes, there are 15766 nodes between the two endpoints. There is no mesh protocol which could find such a long path in acceptable time with acceptable resources. And if the forwarding of the packet over one hop just takes 100 microseconds (the value will probably be much higher in reality) the round-trip time is already over three seconds. An average Internet uplink does the same within milliseconds.

Thus mobile ad-hoc networks always need to be backed up by higher-level, structured networks run by some organization, and at some point (e.g. to cross oceans) they need to rely on wires. Which makes those organizations normal ISPs again. And you gain nothing, because then they would be subject to government regulations again. So there is no "Wireless Mesh technology stack for Internet Freedom". Maybe you can do it in a larger city with a couple thousand friends, but not with whole countries and millions of people. When they pulled the plug in egypt wireless meshes could not have provided any noticeable service to the whole society, just to some small elite with the appropriate tools. And in my definition that is not "freedom".

Building the Technology Stack for Internet Freedom (Gigaom)

Posted Feb 22, 2011 19:28 UTC (Tue) by martinfick (subscriber, #4455) [Link]

You make many good points. You point out how a wireless mesh network could not work between too very distant points. Then, unfortunately, you tie this to a highly overloaded conclusion:

> Thus mobile ad-hoc networks always need to be backed up by higher-level, structured networks run by some organization, and at some point (e.g. to cross oceans) they need to rely on wires. Which makes those organizations normal ISPs again.

Yes, likely wires will be needed, but nothing prevents a non-structured mesh of distributed long haul providers from cooperating using mesh like protocols designed to take advantage of wired optimizations, from servicing many of those needs. Distributed control does not have to mean elimination of all large organizations, it just has to eliminate single large organizations controlling things (i.e. monopolies). The only part of the internet controlled by one organization right now is IPs and DNS allocations. And this centralized control isn't even to solve a technical problem such as routing, that is already distributed by many organizations today. But, yes, it would be better to distribute more of the infrastructure, to have more independent ISPs, more independent infrastructure.

As for allocating IPs and DNS, well, there are many things which could potentially replace the current schemes, the work underway on the DNS part is getting some press for sure. But, IPs are even simpler in many ways, because an IP does not have to be human readable, simple schemes such as the way tor allocates onion addresses can be used to distributed this, without requiring a registry like DNS does.

Mesh networking is making a big splash because it is new. But, decentralizing the internet is actually a much simpler problem than solving mesh networking. It's just that to decentralize the internet (IPs and DNS), you would have to abandon some of the old ways of doing things, and people aren't really willing to just do that without a real pressing need to. It's really a bit like moving to IPv6. Perhaps now is the time to do it, before we make that transition, or as a part of that transition? In fact, I suspect that moving to a non centralized IP allocation scheme might even be simpler than moving to IPv6. Why not simply throw the old allocation scheme away when moving to IPv6?

The appeal of mesh networking is mostly that it is new territory, so there is nothing old to abandon/transition, and people can experiment with it easily. It also makes people question the old assumptions that centralized control is needed, so it gets the press (and mind share). Fixing the old, slightly broken existing infrastructure, that get ways less attention. However, I suspect that eventually, the old assumptions will be seen to be flawed, and some headway will be made to clean up the existing infrastructure.