**6. Conclusions and future works**

18 Will-be-set-by-IN-TECH

**Figure 7.** Per-scenario Power Consumption increase with respect to Idle (in percentage)

enabling the Video Camera.

Power Consumption would be

which gives a 48% overhead on power consumption.

per second of

Old-Gen.

consumption profile of Skype is very different (about 4-5 Watts in average) with and without

Another interesting question that arises from the analysis is, in case of applying these Scenarios in groups, if their power consumption would follow a linear composition rule (thus summing up the values). That is, for example, supposing a composed Usage Scenario *S* that involves a Skype Call, a Web Navigation and a Disk Operation performed simultaneously, their linear composition would give, on our Old-Gen PC, an estimated Power Consumption

*Pidle* + Δ*PS* = 86.81*W* + 21.33*W* = 108.14*W*

introducing a 25% overhead on power consumption. On the New-Gen PC, the estimated

*Pidle* + Δ*PS* = 51.39*W* + 24.90*W* = 76.29*W*

**5.3. Question 3: What is the relationship between usage and power consumption?** Taking a look at the results of the correlation analysis, represented by the third Research Questions, more conclusions can be made. First of all, we can observe that the coefficients related to the New-Gen PC are higher with respect to the Old-Gen PC. This may suggest that as hardware evolves, the software usage is even more significant for determining the power consumption of the system. This assumption is confirmed by Figure 7, where we can observe that the percentage increase of the New-Gen PC is higher, in most cases, with respect to the

However, it is remarkable that, for both machines, the variables that show higher correlation coefficients are undoubtedly those related to CPU Usage and Memory Usage. High This experiment assessed quantitatively the energetic impact of software usage. It consisted in building up common application usage scenarios (e.g.: Skype call, Web Navigation, Word writing) and executing them independently to collect power consumption data. Each single scenario introduced an overhead on power consumption, which may raise up to 20% for recent systems: if their power consumption would follow a linear composition rule, the impact could be even higher.

The relationship between usage and power consumption was also analysed in terms of correlation between resource usage. Although a clear linear relationship did not arise, the analysis showed that some resources drive power consumption more than others, such as memory and CPU usage. Using a precise control over how an application consumes these resources, it can be possible to predict its power consumption, thus including dedicated countermeasures in the Software Design Phase – which, by itself, is the essence of Energy-Aware Programming.

Our experiment also gives us the indication that modern desktop systems, although being more energy efficient in standby and idle states, due to their higher scalability, are even more sensible to the impact of software usage on power consumption. This indicates that research should focus on reducing this impact, as it will be always more significant as time goes by.

Moreover, results set the basis for future works and research projects. A more accurate correlation analysis will be conducted, focusing on the more relevant resources and taking into account also different kinds of relationships (not just linear). Moreover, we will focus our attention on battery-powered mobile devices, where software power consumption is a key issue. Our idea is that re-factoring applications by considering a more efficient resource utilization, the impact of software on power consumption could be easily reduced.
