Home » CurrentHeader, Digital Insight, Wireless

The problem with mesh networks

By 17 July 2010 2 Comments

Despite the fact that “mesh” networks are completely impractical for commercial cellular phone or data applications, it hasn’t stopped the idea from achieving cult status.  Last year I had to debunk Sascha Meinrath of the New America Foundation who claimed that mesh networks were superior and that traditional mobile networks aren’t scalable.  In 2007 I explained why the “One Laptop Per Child” (OLPC) mesh networking scheme is a nonstarter and it didn’t take long for the Mongolian schools to learn the painful lesson that mesh melts down.  The latest one to think that the telecoms are doomed because of a new Smartphone mesh project is open source proponent Eric Raymond.

The fundamental support for mesh networks stems from simple anti-establishment attitudes.  Mesh proponents believe that you could build a large or even nationwide network without any wires or cell tower base stations with just end-point devices and you could completely bypass and obsolete the phone/Internet company.  The mesh cult even comes with the obligatory conspiracy theories that explain why mesh networks aren’t widely deployed due to corporate suppression, but that completely ignore the laws of creative destruction.

Why mesh doesn’t scale

The fundamental problem with mesh networks is that it simply doesn’t scale.  If you wanted to build a network based solely on end-point devices like laptops or Smartphones, it would take at least 64 active and optimally placed end-points just to provide the equivalent coverage of a base station with 8 times the range as shown in figure 1 below.  Real world conditions would likely require many times more end-points and it would be even more difficult to substitute for a long range cellular base station.

Figure 1 – Mesh scales poorly compared to hub-spoke architecture

One might ask how is it fair to give the base station 8 times more range, but that’s the fundamental advantage of a base station not possible with end points.  Base stations usually pick out a fixed and strategically advantageous locations, come with an antennas that’s are many times longer, and have transmitters many times more powerful.  Cell tower base stations are far more powerful than 8 times the range.  End point devices by definition have to move around to serve their human owner and be small enough to fit in their human owner’s pocket which severely limits the size of the antenna and battery.  All these factors make end points horrible makeshift base stations.

History lesson of the OLPC XO:

The OLPC XO computer was a huge failure because the design was tragically weighed down by unrealistic and unnecessary design goals because of its Nicolas Negroponte’s insistence on mesh networks and other technologies.  Negroponte believed that schools and even teachers (which he characterizes as drunks) were unnecessary and that the solution was to “leverage the children themselves” for learning.  Based on this philosophy, his team designed the OLPC XO to function in daylight (spun off to Pixel Qi) so that it didn’t need classrooms, originally designed the XO with a hand crank generator assuming that there would be no electrical infrastructure, and implemented mesh networking to function in the absence of network infrastructure.

The latter two design goals were complete flops.  The outdoor screen seems to have spun off into a viable commercial technology though it remains to be seen how successful it will be.  The generator idea was too difficult to deploy in a sufficiently reliable form factor and the mesh feature simply inflated the cost of the OLPC since there were few off-the-shelf mesh capable radio subsystems and it reduced the XO’s battery life.  The XO had to constantly remain on to act as a mesh relay node which drained the batteries.  A simple $100 Wi-Fi base station would have been a superior solution with a much larger and more reliable coverage area while reducing the cost of every XO laptop.

The true believers of mesh networking even touted Africa’s ability to leap ahead and connect the entire continent of Africa through mesh networks.  The BBC even called this “the Great Leap Forward” which was truly ironic and fitting considering how eerily it matched the Chinese Great Leap Forward of the late 1950s.  They didn’t need any established methods of agriculture or industrialism for they would simply invent their own way and leap beyond the west.  Being born at the end of the Great Leap Forward, I am keenly aware of the tragedy it caused.

Why mesh networks are fundamentally slower

In the real world, mesh networks are overwhelmed by overhead traffic to the point that they melt down with any kind of real user load and the Mongolians learned this the hard way during deployment.  But even without overhead traffic, mesh networks are fundamentally flawed because mesh network are fundamentally many times slower than direct hub-spoke solutions that employ base stations.

We can imagine a cargo fleet of 30 airplanes transporting goods from San Francisco to New York, but having to stop in Las Vegas and Denver Airports. We would have to allocate 10 planes on each leg of the flight rather than all 30 airplanes for one continuous trip from coast to coast. Even ignoring layover delays and the extra time it takes to take off and land, we’re still going to lose 2/3rds of our direct shipping throughput.

Figure 2 below illustrates how each node cuts down capacity because the relay node has to either split the frequency or split the time it can use to forward traffic.  A relay node that receives and transmits at the same time would be forced to use half the radio frequency for receiving and the other half for transmitting.  If the relay node alternates between sending and receiving, it could use all of the radio frequency but it only gets half or 1/3 of the time to transmit.  The bottom line is that a lot of capacity is lost whenever we deploy relay nodes.  The relay design works great when we can’t go direct point-to-point due to structural obstacles like hills, trees, or even just the curvature of the earth, but to deliberately forgo the direct route by eliminating base station infrastructure is just silly.

Figure 2 – Mesh networks lose capacity with each relay node

Mesh might work for military applications

Many proponents of mesh networking often point out how the military is very interested in mesh technology, but they fail to see that military requirements are largely irrelevant for civilian applications.  The military needs end points that can survive attacks on base station infrastructure but this isn’t a concern for civilian networks.  A soldier could carry 40 lbs of communications gear, but civilians balk at anything more than 5 ounces.