Painful lesson in OLPC mesh networking for Mongolians

The Mongolians have had a painful lesson on mesh networking according to the OLPC current events webpage.  Broadcast storms in the overly dense mesh environment along with excessive mDNS broadcast traffic seem to have crippled the Gobi desert experiment.  Here's an excerpt:

We have painfully discovered the limitations of the mesh and current collaborative software in Mongolia, where the convolution of the number of laptops with bugs #5335 (more mDNS traffic than expected) and #5007 (mesh repeats multicast too much) make the perfect storm, which prevents anybody from using the network. We will continue to improve the mesh performance, but clear guidelines are needed as to what network infrastructure to deploy under what conditions. Once a certain density of students is exceeded, a wired backbone and conventional access points will be required.

The limitations of mesh topology are well known in the wireless engineering community and I've raised the issue and pointed out the limitations last September.  Each mesh hop you add increases the propagation delay as well as multiply the radio traffic and congestion.  Performance on a mesh network is fundamentally many times slower than a non-mesh network and when the density gets high enough, the system simply breaks down.

When on a tight budget, I had always recommended the usage of a cheap $60 router running open source DD-WRT would have sufficed and you get a free router with it which you need for IP sharing anyways.  The addition of a high-powered antenna would allow the access point to hear distant signals from faint clients and it will amplify the broadcast signal.  A simple in-door $26 9 dBi antenna placed up high can easily cover a small school.  A $60 12 dBi outdoor antenna positioned on the roof would easily cover an entire campus.  If you put two centralized Access Points and large antennas on channel 1 and 11 (avoid adjacent channels because of channel bleeding) in the 2.4 GHz spectrum, you can load balance and have redundancy if one set of AP/antenna fails.

My fellow blogger and teacher Chris Dawson feels that the ability to do peer-to-peer collaboration with or without an Access Point has great potential.  But peer-to-peer wireless collaboration could have been done with regular ad hoc networking technology without the expense or problems of a full 802.11s mesh implementation.

The inclusion of full 802.11s stack has been challenging.  The need for a radio system that stays on and continues to forward packets even while the laptop is off added unnecessary expenditure to the OLPC XO and it unnecessarily drains the laptop batteries.  When you multiply this expense and complexity across all the clients and realize that the wireless access point comes free with the router, it becomes clear that this may not have been the best design decision.