Information Technology Reference
In-Depth Information
CSV File with
WSN data
(1)
WSN Sensor Nodes Database
(2)
Fifty Days Sample for
selected month
(3)
FPM input Data Retrieves
(4)
FPM Execution
(5)
Results
Fig. 1. Frost Detection Module
5 Performance Estimation Models
In the present Section we present our models to estimate the performance of EC2
instances for processing frost prediction applications. The methodology used to
construct the models is the following: first, we execute the frost prediction ap-
plication in each instance to obtain empirical results of performance indicators
(execution time and economic cost). Then, we use polynomial expressions and
empirical results to generate the performance models. Next, we extract conclu-
sions about the accuracy of the proposed models. Finally we select the most
suitable instance for frost prediction through a comparison in a typical use case.
5.1 Frost Prediction Application Execution
The execution consists of running the frost prediction application and measure
the execution time. In order to extract the average value of the execution time,
the procedure is repeated four times for different number of sensor nodes (from
10 to 1000) in each instance. Finally we use the average execution time and the
pricing list of Amazon to calculate the economic cost required to execute the
application.
We have considered five test scenarios, one for each instance types to model
(see Table 1). Each row in the Table 1 represents the different instance types, i.e.,
t1.micro, m1.small, m1.large, m1.xlarge and c3.xlarge, and each column indicates
the instance characteristics, i.e., number of virtual CPUs (vCPUs), Amazon EC2
Compute Unit (ECU), Memory (expressed in GBytes) and Instance Pricing.
Regarding the Amazon's pricing model used in our work, we use the
“on demand”
