[Olpc-open] OLPC and Wireless Sensor Networks: what can be learnt?

[Tom Parker]

(Apologies to anyone who's seen this before. I sent it to 
research at laptop.org about 3 weeks ago, but that mailing list appears to 
be a bit on the dead side, so here's a second attempt...)

Summary: Wireless Sensor Networks (WSNs) have managed to do a number of 
things not considered for traditionally networked systems (physical 
localisation; routing designed for wireless; etc). These could also be 
used for the OLPC project with the current hardware design, but have not 
been done so far.

Wireless Sensor Networks are based upon the idea of many small computing 
nodes designed for data acquistion, processing and retrieval via radio. 
To some extent, it is a daughter field from Mobile Ad-Hoc Networking, 
but with a focus on vastly reduced resource budgets (energy, processing, 
memory, radio bandwidth). Typical hardware for sensor networks has a 
single-digit MHz 8-bit microprocessor, 4-10 KB of RAM, and a total 
energy budget similar to a pair of AA batteries. Typical runtime with no 
power management is 4 days, but usual requirements are for 2-6 months, 
so we've had to try and be a bit smarter with our power management. The 
Wikipedia page [1] is pretty good for a wider view of the field.

Despite all of these limitations, or perhaps because of them, a number 
of innovative solutions and new capabilities have been created for these 
devices, and I feel that some of this innovation should be fed back to 
other wireless networks (i.e. OLPC).

* One of these areas is physical localisation. Much effort has been 
expended in attempting to work out where we have placed our sensor 
nodes, as otherwise our data acquistion is of limited use. Some of the 
solutions involve additional hardware (ultrasound emitters for example), 
but a great deal of work has been done in attempting to do smart things 
with the inaccurate data from Radio Signal Strength information, which 
is also potentially available to the OLPC computers. Imagine a 
Neighbourhood view with information about where your friends are, and 
who's moving around.

* Another area is wireless-aware routing. The Guesswork protocol [2] for 
  example, is designed around the idea of broadcast as the basic 
primitive for routing vs. the unicast-based routing generally thought 
about in most scenarios, as unicast is inefficient for unreliable radio 
links. Additionally, the protocol can be used to do energy-efficient 
multicast which would be useful for package dissemination. A critical 
note is that the protocol will not really work in wired networks, as it 
is designed around broadcast messages and overhearing of packets. For 
wireless networks however, it is perfect. An implementation of this 
protocol has already been achieved for sensor nodes with the 
specifications above (i.e. minimal memory/processor usage).

Couple of other things of note: we're doing this with hardware less 
advanced than your radio chip, our solutions tend to be fully 
distributed, and our software is designed to work well in networks with 
*many* nodes e.g. a school of OLPC users.

So, why am I writing this? Well, to be honest, I'm fishing. I'm in the 
process of finishing my PhD (estimate finished ~September, with thesis 
subject being how we can rethink our abstractions and mental models to 
create better protocols for sensor networks, with a side note of "this 
can be used elsewhere"), and I'm hunting for my next project. In 
addition to the Guesswork protocol mentioned above, I also have 
published work on smart RSSI-based localisation protocols, and other 
areas of Sensor Networks. My full publications can be found at [3]. I'm 
searching ideally for post-doc opportunities (although commercial R&D 
may be considered) to explore some of these possibilities, and see what 
can be done about doing some really smart things with the OLPC hardware. 
Anywhere in an English-speaking country is fine by me (3.5 years in the 
Netherlands have taught me I have no talent for human languages).

Thanks for your time. Any takers?

Tom Parker
-- 
palfrey at tevp.net
PDS Group at TU Delft, Netherlands

1) http://en.wikipedia.org/wiki/Wireless_sensor_network
2) http://tevp.net/papers/guesswork-mass.pdf
3) http://tevp.net/papers/