Métodos de análisis sísmico según normativa

Building on the concept of conservativeness (i.e., the tendency to select a lower MCS/TXP in the transition regions), we explore whether there is any correlation with the energy efficiency achieved by a certain algorithm and this tendency. For that purpose, we first define a proper metric.

In the first place, we define the normalised average MCS as the area under the curve in Figure7.4(top) normalised by the total simulation time and the maximum MCS:

MCSˆ = ¹

max(MCS) ·tsim

∫ t_sim

0 MCS(t)dt (7.1)

where t_simis the simulation time and max(MCS)is 54 Mbps in our case. The same concept can be applied to the TXP:

TXPˆ = ¹

max(TXP) ·tsim

∫ _tsim

0 TXP(t)dt (7.2)

where max(_TXP)is 17 dBm in our case. Both_{MCS and ˆ}ˆ _{TXP are} unitless scores between 0 and 1, and lower values mean a more con-servative algorithm. Therefore, we can define a Concon-servativeness Index (CI) as the inverse of the product of both scores:

CI= ¹

MCSˆ·TXPˆ (7.3)

where CI>1.

We computed the CI¹¹for each device and run, and the final re- ¹¹It must be taken into account that the CI is not suitable for comparing any algorithm. For instance, in an extreme case, an “algorithm” could select 6 Mbps and 0 dBm always, resulting in a very low CI, but a very bad performance at the same time. The CI should only be used for comparing similarly performant algorithms, as it is the case in our study given the results shown in Figures7.2and7.3.

sults are depicted in Figure7.5as the average CI across different runs vs. the median energy efficiency in Figure7.3.

The results in Figure7.5show a positive non-linear relationship between the CI of an algorithm and the energy efficiency achieved for all the devices considered. MP is the algorithm with the lowest CI, which is in consonance with its aggressiveness (i.e., frequent jumps between MCS/TXP values, as shown in Figures7.4, and the goodput

HTC Legend Linksys Raspberry Pi Samsung Galaxy Note Soekris

22 23 24 258.5 9.0 9.5 10.0 11.5 12.0 12.5 13.0 13.5 22 23 24 7.0 7.5 8.0 8.5 select lower MCS and TXP) and energy effienciency per simulated device.

achieved was also the lowest, as depicted in Figure7.2. On the other hand, PARF, PRCS and RRPAA achieved a similar performance in terms of goodput, but the ones with the most conservative behaviour (PRCS and RRPAA, as it can be seen in Figures7.4) also achieve both the highest CI and energy efficiency.

This result evidences that the performance gaps uncovered by Figure6.4under optimal conditions have also an impact in real-world RA-TPC algorithms. Therefore, we confirm that this issue must be taken into account in the design of more energy-efficient rate and transmission power control algorithms.

7.5 Summary

We have extended¹²our results from Chapter6regarding the role of ¹²I. Ucar, C. Donato, P. Serrano, A. Garcia-Saavedra, A. Azcorra, and A. Banchs. On the energy efficiency of rate and transmission power control in 802.11. Computer Communications, 117:

164–174, Feb. 2018a. ISSN 0140-3664.

d o i: 10.1016/j.comcom.2017.07.002 Ucar et al. [2018a]

conservativeness at mode transitions in achieving better energy effi-ciency in RA-TPC algorithms. We have developed a metric to com-pare algorithms, and we have assessed the performance of four state-of-the-art schemes through simulation. We have demonstrated that certain conservativeness can resolve the trade-off between through-put and energy efficiency optimality, thus making a difference for properly designed energy-aware algorithms.

Further research is needed to develop proper heuristics to leverage these findings. In particular, the downwards direction, as described in Section6.3.2, is the most challenging, because it requires predicting the evolution of the channel state.

Simulation frameworks are important tools for the analysis and design of communication networks and protocols, but they can result extremely costly and/or complex (for the case of very specialised tools), or too naive and lacking proper features and support (for the case of ad-hoc tools). Our own research experience in the previous chapter has pointed us towards the need for new tools sitting be-tween these two paradigms, supporting the efficient simulation of relatively complex scenarios at a low implementation cost.

In this chapter, we introduce a recent event-driven simulation package,simmer, and show its applicability in fast prototyping.

simmersits between the above two complexity extremes, and com-bines a number of features that supports, among others, versatility and repeatability. More specifically, some of the key advantages of simmerare as follows:

• It is based on the very popular R programming language¹, which ¹R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria, 2017. URLhttps:

//www.R-project.org/

R Core Team [2017]

benefits from a large community of users and contributors, but also natively supports the analysis of results via the many R statis-tical and visualization packages.

• The code has been peer-reviewed, and it is an official package², ₂

I. Ucarand B. Smeets. simmer: Discrete-Event Simulation for R, 2018. URL https://CRAN.R-project.org/package=

simmer. R package version 3.7.0 Ucar and Smeets [2018]

with numerous examples readily available, potentially supported by a notable user population.

• In addition to its ease of use and versatility, its code is partially op-timised for speed, and therefore it can simulate relatively complex scenarios under reasonable times.

In the following, we first describe the simulation core design and its architecture. Then we provide a quick overview ofsimmerand its key features. Finally, we showcase the versatility ofsimmerto easily model a Massive Internet-of-Things (IoT) scenario where thousands of metering devices share the same channel. Here, we analyse the impact of access parameters on performance, with a particular in-terest in the energy required to deliver the information, which will ultimately impact the lifetime of devices running on batteries.