Fig. 6. Concert area, 5:00 pm: complementary CDF for the infection rate with the enabling radius equal to 100 m.
By comparing this value with the total number of infected nodes (Fig. 5-b), we have that the ratio between the
infected and the enabling nodes is always greater than 10, thus proving that the alarm diffusion is mainly due to ad hoc
communications. However, the results also show that the infected node number is deeply affected by the number of active
nodes and that only a fraction (between 5% and 40%) of the devices are able to receive the alarm in a timely fashion.
To exclude from our analysis nodes that in the considered time interval are far away from the emergency location
(the furthest are roughly 100 km distant), in the following two figures we limit our attention to the infection rate for the
target areas. With regard to the smaller area (Fig. 5-c), we have that, at the beginning of the alarm spread, the infection rates
increase smoothly, while after ten minutes all the curves start exhibiting a steep slope. Although less evident, this behavior
is still present in the results for the larger area (Fig. 5-d). With reference to the smaller target area, the fraction of infected
nodes is remarkable when a significant amount of nodes cooperate in alarm spreading, reaching after 30 min on average
50% of infected nodes when half of the nodes are active and a value greater than 70% in the most favorable case. In most
cases, rates are roughly halved for the larger target area.
Finally, in Fig. 6 we analyze the distribution of the infection rate among the different simulations by means of the
complementary cumulative distribution function (CDF). The complementary CDF for a random variable X denotes the
probability P(X ≥ x). The results show approximate normal distributions. Moreover, they show that when only a fraction of
the nodes are involved in alarm forwarding, the infection rates deeply depend on how these nodes are selected. For instance,
if we consider the scenarios in which roughly half of the nodes cooperate for alarm diffusion (300 and 400 active nodes
respectively), we have that after 30 min the infection rate varies between 20% and 70% for the smaller area and between
10% and 40% for the larger one.
In the second scenario, we consider again an alarm originating at the concert area at 05:00 pm, but with an enabling
radius equal to 200 m. Clearly, the larger the enabling area is, the higher the infection rates are, as shown by Figs. 7 and 8.
However, by comparing Fig. 5 withFigs. 6 and 7 withFig. 8, respectively, it is easy to see that the behavior of the considered
metrics does not change significantly when the enabling radius is doubled.
In the third scenario, we consider an alarm originating at the shopping center at 05:00 pm, and we compare the results
for the infection rates with those obtained for the first scenario. Looking at Fig. 9, we observe that, not only are the infection
rate values after 30 min similar to those obtained for the concert area scenario, but also the behavior of the curves is quite
similar. This behavior is reasonable since the two originating locations are close to each other and thus both the mobility
patterns and the human density are quite similar.
Therefore, in the fourth scenario we consider an alarm originating at the concert area at 09:00 pm. Here we have that
all the infection rate values are higher than those obtained in the previous scenarios (Fig. 10). Moreover, we have that the
infection rates increase rapidly at the beginning of the alarm spread. The main reason for this is that the characteristics of the
mobility patterns of this scenario slightly differ from those of the previous two, mainly because people congregate around
the concert area at 09:00 pm to attend the concert.