OLPC Mesh Networking Tester Explains How It Works

An anonymous reader writes "James Cameron is an engineer working on the OLPC project, specifically testing the wireless network capabilities of the OLPC XO laptop. Cameron lives in a small town called Tooraweenah in a remote region of the Australian outback. There is little noise in the spectrum in the area, so it's perfect for testing the wireless networking capabilities of the XO as it mirrors the kind of rural, spacious environment the XO is intended to be deployed in. Cameron breaks down exactly how the OLPC XO's mesh networking works, including the cheap US$35 solar powered mesh nodes that can be mounted on top of a tree to further the network's reach. Testing in the Australian outback, Cameron discovered that the range of the XO could go up to 1.6km 'quite easily' at 1.5m above ground. 'Assuming a range of 1.6km holds true, (the mathematical formula for area of a circle) Pi R squared tells us one well placed mesh node will cover up to eight square kilometers.' The article also includes numerous pictures of the mesh nodes and testing of the XO."

The ugly truth about mesh networks

[Anonymous Coward]

In a wireless mesh network, 99.9% of which are implemented using half-duplex transceivers, once past your immediately adjacent nodes, for each additional hop away you halve the bandwidth and double the latency. No way around that, and in a large area mesh (think muni wifi here) you must re-insert an additional backbone feed about every 4 to 8 hops into the mesh either by landline or by point-to-point wireless bridges, else performance across the mesh gets abysmally poor, very quickly.

Dense meshes just don't work very well, they implode upon themselves. Very sparse meshes, such as used in the battlefield by our military, of perhaps in remote areas like the Aussie Outback as mentioned in the FA, are ideal applications of a wireless mesh network, but all the folks who think they can make a successful commercial venture with a wireless mesh in a dense urban or suburban environment are in for a rude awakening if they drink too much of the Koolaid hype that many of the consumer-grade hardware vendors are trying to push.

open mesh

[qw0ntum]

For those who want to build their own mesh, check out the open-source ROBIN [blogin.it] project. They are building a complete plug-and-play mesh networking package. they are even configured to automatically connect to the Open Mesh Dashboard [open-mesh.com] so you can manage your network. Open Mesh will start selling pre-flashed nodes this week at their site.

Re:Why not release schematics and other info?

[legutierr]

The OLPC wiki [laptop.org] is very extensive and growing, and is a great starting-point for making contact with the core OLPC team (including senior management at OLPC, who are surprisingly accessible). If there isn't an article specifically addressing a person's questions, the wiki at least provides a place where questions can be posted with a reasonable expectation of getting an answer. In addition, there are instructions on how to join OLPC mailing lists and IRC, where you can communicate directly with the team.

Nevertheless, I will say that much of what is on the wiki is oriented to software or organization building, and that info on OLPC hardware seems to get the short-shrift. There is a *short* article discussing the wireless repeater here [laptop.org], which links to the manufacturer's page here [marvell.com]. I'm sure that contributions to the wiki in this area would be appreciated.

Re:The ugly truth about mesh networks

[Agripa]

I will lose my membership for not knowing this, but aren't there 11 channels (not sure how many non-overlapping channels though). How can 4 machines DoS every wifi channel when they aren't even doing anything!

In the US, the 2.4 GHz ISM band has 11 channels spaced 5 MHz apart. 802.11b and 802.11g require 25 MHz of separation to prevent interferance which limits the non overlapping channels to 1, 6, and 11. 802.11n and many 802.11g systems support double channel widths of 40 MHz which limits the 2.4 GHz ISM band to just one non overlapping channel.

The 5.0 GHz band used for 802.11a and for some 802.11n radios has 19 20 MHz channels alleviating much of the congestion problem at the expense of cost and using a higher frequency.

http://en.wikipedia.org/wiki/List_of_WLAN_channels [wikipedia.org]

Re:The ugly truth about mesh networks

[jeffstar]

In fact the rate at which latency increases and bandwidth decreases in a mesh network has been shown to be 1.68^N where N is the number of hops (Piyush Gupta, Robert Gray, P.R. Kumar An Experimental Law for Ad Hoc Networks, May 2001). I have measured this myself in an 8 hop half duplex wireless network.

That is only for data being sent straight through the mesh. When clients connect directly to the mesh nodes on the same radio that is used for sending and receiving mesh traffic then there is even less radio time to go around and things slow down even more.

There is a company called Mesh Dynamics [meshdynamics.com] which sells mesh nodes with three radios [meshdynamics.com]: one for send, one for receive and another for clients to connect to.

I recall they have modified or extended in some way the linux kernel to have one radio dedicated for sending and another for receiving. This is something I am interested in doing as well but haven't seen any kernel modules or extensions to allow it thus far.