Information Technology Reference
In-Depth Information
pricing model. It is noteworthy that in this paper we do not make an analysis
of the accuracy of the minimum temperature predicted by the frost detection
application. However (and based in our experience with agronomists) we can
arm that an error of +/- 1.5 celsius degrees is an acceptable error value to
predict frosts, and the used FPM meets this requirement.
Tabl e 1 . Test Scenarios
Amazon EC2 Instance vCPUs
ECU
Memory (GBytes) Pricing on demand (U$S)
t1.micro
1
variable
0.615
0.020
m1.small
1
1
1.7
0.047
m1.large
2
4
7.5
0.190
m1.xlarge
4
8
15
0.379
c3.xlarge
4
14
7.5
0.239
The application execution allows to obtain empirical performance results in
each EC2 instance. Figure 2a shows the execution time versus the number of pro-
cessed sensor nodes for the scenarios considered. Figure 2b details the economic
cost versus the number of processed sensor nodes.
(a) Execution Times.
(b) Execution Costs
Fig. 2. Empirical Results.
From the Figure 2a can be observed that m1.large is the instance which have
achieved the shorter execution times for the frost prediction application. In ad-
dition, it can be seen that up to 200 sensor nodes processed, the performance of
m1.large is notable. Next, the performance of m1.large becomes similar to the
m1.xlarge and c3.xlarge.
Furthermore, results show that for multiprocessor machines as m1.large and
c3.xlarge, the processing times decrease for 30 and 40 sensor nodes, respec-
tively. Regarding the observed decrease, the decrease in m1.large instance is
 
Search WWH ::




Custom Search