Information Technology Reference
In-Depth Information
Regarding economics costs, a particular case is when the execution times are
close to an hour. In this situation, if the execution time calculated by the model
is longer than one hour, the costs predicted by the model will be twice than
those empirical. This is because the price of EC2 instances is per hour of use.
Likewise, if the model predicts less time than one our, the cost calculated by the
model would be half than empirical costs. However, when the proposed model
is used this situation does not happen, so we can say that the accuracy of the
models regarding economic cost is suitable.
5.3 Instance Performance Comparison in Typical Use Case
In order to select which is the instance that presents the best performance for
running frost prediction applications, in this subsection we present a comparison
in a typical use case of frost prediction.
The typical use case consists of WSNs deployed in different farms in the
Province of Mendoza. The prediction was made for one day of July because it is
one of the months of frost season, which begins in April and ends in October.
The frost prediction application runs on the Cloud and predicts the mini-
mum temperature, which allows to alert the agronomist engineer. Finally, the
agronomist decides if the guard procedure against frosts must be conducted.
Regarding frosts guard procedure, it consists in moving the staff to the farm
and wait the decision of the specialist, who in turn decide - on the basis of data
collected in real time - the activation of the defense system (heaters, surface
irrigation, sprinklers, etc.).
Another aspect to consider is that the processing of data has time constraints.
This problem arises because in July the logistic of the defense system against
frosts requires that the farm staff must be alerted before 22:00 h. However,
the frost prediction application needs the
T
0
temperature which is registered
in July of Mendoza province about the 21:00 h. According to the above men-
tioned reasons we can conclude that the maximum execution time allowed for
the application must be less or equal to one hour.
Table 3 shows the number of nodes processed by each EC2 instance model for
one hour and the execution cost.
Tabl e 3 . Processed Sensor Nodes in Maximum Execution Time
Amazon EC2 Instance Nodes Economic Cost (U$S)
t1.micro
78
0.020
m1.small
324
0.047
m1.large
841
0.190
m1.xlarge
632
0.379
c3.xlarge
722
0.239
Results showed that the most suitable machine for this application type is the
instance m1.large. The reason is because the m1.large is the machine that can
