Meeting 12 Nov 2010
We discussed progress since last week, and I showed the plots from the MIT-Reality dataset, and decided that it probably is OK for determining whether nodes are co-located or not, and give some extra information about movements. I said that I would generate a graph of people connected during the same time period to the same cell tower.
Pádraig suggested a simple policy for routing, based on ranking of locations and consequently people that visit them, where a persons rank is based on the rank of the towers (places) he visits, and how many different towers he visits, making a person that moves around to different, highly ranked towers, more important for routing that someone that only visits low ranked towers, or perhaps even someone who only visits one highly ranked tower.
Davide said that he might have made some notes about this in the past and would look them up, and Pádraig said that it could be calculated within some probablistic framework. I noted, that this sort of scheme would really be routing towards the center of the ranking network, which Pádraig suggested was because it was routing based on the centrality of the network. But we decided that it is worth using this to at least test this idea out against a random routing protocol.
With regards the actual implementation, Pádraig explained his idea that we could do the ranking (at least to start with) before the simulation starts, and simple gives nodes fixed ranked positions, meaning that we don’t actually have to change any of the simulator code, the rank is just a property of the node itself. So, a routing policy is as follows:
on meeting another node:
foreach message in buffer
if node is destination node
pass message
else if node has a higher rank and visits a
location that the destination node does
pass message
else
keep message
The idea being the the protocol uses the existing contact events to drive communications, but uses the fixed location ranking to form the routing policy.
Pádraig also said that we really need to have a conference target in mind, and we couldn’t really think of a suitable one, Davide suggested Ubicomp, but it has a low acceptance rate.
So, for next time, I will have generated a graph of connected individuals based on connecting to cell towers during the same time period, made sure I can generate the same results as Graham, generated a ranking for individuals based on cell towers, developed a routing scheme based on this, and hopefully identified conference targets.
UPDATE:
Davide suggested the following calculation for determining the popularity of users, which can be used as a metric for routing.
Popularity of user [latex]i = \sum_j \tau_{ij} ( p_j + c )[/latex]
where [latex]\tau_{ij}[/latex] is the time (or number of times) user [latex]i[/latex] has visited the tower [latex]j[/latex], [latex]p_j[/latex] is the popularity of tower [latex]j[/latex] and [latex]c[/latex] is a parameter which tunes the importance of user mobility
Pádraig suggested that [latex]c[/latex] should be 0.