Fig. 7. Concert area, 5:00 pm: average of 50 simulations with the enabling radius equal to 200 m.
To confirm this reasoning, in the last figure (Fig. 11) we analyze the distribution of the number of nodes infected by
each active node (at the end of the alarm spread) by means of the complementary CDF with logarithmic scale. Here, the
complementary CDF for a value x denotes the probability that an active node is responsible for having infected more than x
nodes.
We found that in general a few active nodes are responsible for most of the information spreading. In fact, for both the
scenarios the distributions approximate a power-law (straight lines in logarithmic scale), confirming the results obtained in
smaller experiments. Moreover, by comparing the two scenarios we have that during the 09:00 pm scenario (Fig. 11-b) the
distributions of infected (and thus met) nodes are higher than those of the 05:00 pm scenario (Fig. 11-a). Finally, when the
number of active nodes is smaller, the distributions are rather flat, because each active node has the possibility of infecting
many passive nodes. This is in contrast with what happens for a larger number of active nodes, where the distribution
decreases more sharply.
5. Experimental limitations
Only one large scale event happened in the Boston metropolitan area during the considered time interval. We plan to
collect data for other events in other cities in order to compare performance based on the type of event.
The number of nodes considered is a subset of the real number of mobile phone users at the event. This limitation is
derived from the requirement to select users for which location information is sufficiently frequent to infer traces. This may
create biases in the selected users. We expect that increasing the number of users would improve routing performance.
Finally, it is important to note that the large scale event analyzed did not involve any emergency situation. While we
agree that human behavior might change in the case of emergencies [18], we believe that our analysis is still of interest as it
relies on real human spatial distribution over large areas, and simulates communication opportunities during special events
not involving large changes in mobility patterns.