Scheduling means timing actions in a proper way. Mobile systems undergo such decisions continuously. For example, when to transmit the next packet or when to switch to executing another process. Traditionally, computing systems have strived for maximum performance or throughput of a system when designing different kinds of scheduling mechanisms. Considering energy consumption provides a different perspective.

How scheduling saves energy

The first motivation for considering scheduling for the sake of energy efficiency is that smartphones and their subsystems are not perfectly power proportional, sometimes very far from it. In a perfectly power-proportional system the power draw would linearly scale with the workload throughout the entire range of possible workloads.

To understand the benefits of energy-aware scheduling, let us consider an example of wireless communication with a smartphone. Wireless communication typically provides better energy utility with a higher data rate. In other words, a higher data rate leads to fewer joules consumed per bit transmitted or received. Because of this, it would be more energy efficient to schedule multiple lower-rate data flows in a maximally overlapping manner than to schedule them individually.

Generally speaking, it is beneficial to exercise the above kind of “race-to-sleep” scheduling policy on subsystems whose power consumption scales sublinearly with the workload. WNIs exhibit such behavior because of their inherent tail energy, as shown by Pathak et al. [1]. The main challenge is to find the optimal scheduling algorithm that takes the possible constraints of the particular subsystem into account. Scheduling algorithms in general have been studied for several decades.

We have so far learned the principles of how smartphones and their different components consume energy. We have also looked at the power-management mechanisms that are employed with the hardware components. Examples of such mechanisms include DVFS to scale the CPU power draw with the workload, and the use of sleep state and discontinuous reception to optimize the energy consumption due to wireless communication.

In this part, we take a look at a set of mechanisms to reduce energy consumption that are built on top of the existing power-management techniques. In this chapter, we first describe the nature and necessity of these mechanisms on a high level after which we briefly introduce the different classes of solutions.

Underlying power-management techniques alone are not enough

Although almost all the hardware contained in modern smartphones has certain kinds of in-built power-management mechanism, usually the whole device still consumes more energy than would strictly be necessary unless another layer of optimization mechanisms is employed. That slightly counter-intuitive observation stems from three main observations:

1. The hardware components are not perfectly power proportional to the offered load.

2. The energy utility of the different subsystems, especially of wireless communication, is context dependent.

3. The underlying hardware power-management mechanisms are rarely optimal for the applications being used.

To improve the energy efficiency of a smartphone, a reliable way to measure the energy consumption of the phone is required. A reliable energy measurement method gives consistent results under the same usage scenario, and responds quickly to variations in energy use. Any energy-efficiency improvements, and energy performance with different workloads, can then be measured with the energy-measurement method.

This chapter discusses energy measurement. We discuss the nature of energy and how it must be measured, and the case of the smartphone and its subsystems as the object of study. We explain direct hardware measurement of power, profile-based software power estimation, building such profiles from hardware measurements, and conclude with a hands-on example.

Hardware- and software-based energy measurement

Draining energy from a battery is a continuous chemical process. The only way to be sure how much energy is being consumed is to measure it constantly as it is being drained. However, doing this affects the use of energy in different ways depending on the measurement method used.

There are two main types of energy measurement: hardware-based and software-based. Hardware-based measurement uses physical tools like power meters that are wired to the smartphone hardware and/or the battery interfaces. Software-based measurement runs on the smartphone energy-profiling software that records the readings of platform-specific built-in indicators of energy use. Note that the software used for measurement also drains the smartphone battery.

To develop energy-efficient techniques, the first step is to understand how energy is consumed on a mobile device. A mobile device consists of hardware components, such as microprocessors, wireless network interfaces, storage, cameras and a touchscreen, and software running on top of these hardware components. Lower-power serial buses facilitate the communication between the internal system components. These hardware components are the actual energy consumers.

Smartphone and mobile device power optimization happens on multiple levels:

1. • Silicon-level, in which the transistor capacitance and the chip design affect the energy efficiency. Higher capacitance requires the transistors to do more work.

2. • SoC-level, in which multiple power/voltage/clock domains can be used to support granular power management with the help of software. In addition, dynamic voltage and frequency scaling (DVFS) is used to dynamically adjust both the voltage and frequency to meet the given energy and performance level.

3. • Software-level, in which various power managers monitor and control the energy and power settings. A high-level framework is needed to perform system-wide tuning and optimization.

There are several choices that contribute critically to the overall efficiency of a mobile device, for instance from an energy-consumption viewpoint. They are the SoC including the CPU, display technology, communications technology, and the OS. The system-level power management is coordinated by the OS. In this chapter, we survey these crucial components and examine their energy consumption.

Mobile operating systems are a relatively new development in the computing world. Mostly they are based on some existing older OS, typically written for stationary computers and laptops. For example, Android is based on the Linux kernel, iOS on the Mac OS X kernel, and Windows Phone on the Windows NT kernel. The mobile OS borrows many features from the desktop world; however, it needs to meet the demands of the mobile environment for communication needs, positioning of the device, power challenges caused by mobile operation with rechargeable batteries, and smaller size for the convenience and comfort of users. The OS is responsible for overseeing and managing the hardware and software components on the smartphone and it is responsible for the system-level power management.

In this chapter, we examine four current state-of-the-art smartphone OSs – iOS, Windows Phone, Firefox OS, and Android – as well as a number of energy-conserving research prototypes. We focus on the current generation of mobile OSs given our focus on smartphones. Classical examples of earlier feature phone OSs include the Symbian OS and Windows Mobile [1]. After examining the four smartphone OSs, we compare their features and properties.

Overview

Power management is an integral part of an OS that operates on multiple levels from drivers to applications. As the functionalities and features of mobile devices are constantly growing so is the demand for power.

Smartphones have evolved in leaps and bounds during the last decade and this evolution is continuing with new software and hardware capabilities as well as new application domains, such as augmented reality and the Internet of Things (IoT). In this chapter we briefly examine future smartphone trends from the perspective of energy consumption. We conclude that radical disruptions are not likely in the near future, keeping energy a first-class resource for mobile devices and warranting new solutions to maintain and improve device battery life.

Future smartphone

The smartphone has become the essential instrument for our daily lives, supporting a plethora of usage scenarios from the workplace to the home and leisure. Today's smartphone is a small-factor, high-performance computer that can offer graphics resolution and power on a par with the previous generation of game consoles. The communications capabilities of smartphones have evolved tremendously over the past decade and now high-speed, low-latency 4G and beyond networks, Wi-Fi, and local area networks are ubiquitously supported. The sensing capabilities have improved dramatically with GPS, acceleration, light, and temperature to give some examples of onboard sensors and the sensing capabilities include complex gestures and voice.

In addition to the hardware capabilities, the mobile-application ecosystem has had stellar success in recent years since iOS, Android, and Windows Phone were launched. Application developers have used the mobile platform APIs and the hardware capabilities of smartphones to create new innovative applications that in many cases combine sensors and communications capabilities in unexpected ways.

In this chapter, we continue to look at the different ways to reduce the energy consumed by wireless communication. Our focus is on how to take advantage of the fact that modern smartphones include many different wireless technologies integrated under the hood and can even switch seamlessly between some of them.

How using multiple WNIs saves energy

Smartphones today contain many different radio technologies including Wi-Fi, Bluetooth, BLE, and cellular radios. As we learned in Section 7.3, using different WNIs can cause quite different amounts of energy to be consumed. More importantly, the energy utility of the different technologies vary substantially. For this reason, opportunities to save energy arise by using the WNIs wisely.

Recall from Section 7.3 that energy is consumed in non-connected and connected modes. The former essentially means discovering an AP or another device in order to establish communication, while the actual data transfer happens in the latter mode. The amount of energy spent in such a discovery process can account for a large part of the total amount of energy consumed. Fortunately, the differences in the energy consumption between the different kinds of radio can be used in that process. In addition, keeping multiple radios continuously powered on in the smartphone is usually unnecessary. In most cases, it is enough to have one radio active so that the phone remains reachable at all times to be able to receive phone calls or incoming messages pushed by mobile services.

In this chapter we give an overview of smartphone batteries and their modeling. Battery models are important when the remaining battery capacity needs to be estimated. In this chapter, we give an overview of smartphone Li-Po batteries, their static and dynamic parameters and characteristics, and charging technology, and then consider techniques for assessing and modeling a battery SOC.

Overview

Figure 3.1 presents an overview of the Li-Ion battery charging and discharging process. Current is carried by lithium ions (Li+) that move from negative (anode) to positive (cathode) electrodes during discharging. The lithium ions move in the reverse direction when charging the battery. The ions move through the non-aqueous electrolyte and the separator. Applying a charge places the battery in a closed-circuit voltage, in which the voltage behavior is set by the internal battery resistance. Charging and discharging distorts the battery and it can take up to 24 hours for the battery to stabilize.

Battery life is proportional to the active reaction sites across the cathode. When the discharge current is low, inactive reaction sites are uniformly distributed over the volume of the cathode. When the discharge current is high, the volume of inactive sites at the outer surface of the cathode is large, causing active sites to become unreachable. As a consequence, the battery capacity is reduced at high discharge rates. When a high current is drawn from the battery, the diffusion rate cannot match the rate at which ions are being absorbed at the cathode.

We now turn our attention from individual optimization techniques to applications. We investigate a few different cases where the principles and techniques we have learned earlier can be applied in mobile applications.

We first look at a specific application, namely video streaming, which is one of the most important internet applications today, from the mobile internet's perspective. We first study the way that video streaming consumes energy and illustrate that through measurement results from real systems. We then cover different strategies that can be used to save energy in video streaming. It turns out that there are a few things that need to be taken into account when applying generic energy-saving techniques to mobile video streaming.

The next two examples are not really specific applications but rather integral parts of many applications and, therefore, they represent extremely important cross-application scenarios. The first of those is sensing. Sensing is a hot research topic at the moment and it is expected to become a very important part of smartphone applications. We study ways to reduce energy consumption with applications that require different kinds of sensing by exploring separately each category of sensors included in modern smartphones. We focus on two well-established techniques: sensor selection and duty cycling.

The second cross-application energy optimization scenario is security. We look at the energy overhead caused by security protocols and algorithms, based on measurement studies. Then, we discuss whether and when it is possible to find a tradeoff between the level of security and energy consumption.

Stationary distributions of queueing systems with general arrivals and service-time distributions are difficult to compute. In this chapter, we will develop techniques to understand the behavior of such systems in certain asymptotic regimes. We consider two such regimes: the heavy-traffic regime and the large-deviations regime. The analysis in the heavy-traffic regime can be thought of as the analog of the central limit theorem for random variables, and the analysis of the large-deviations regime can be thought of as the analog of the Chernoff bound. Traditionally, heavy-traffic analysis is performed by scaling arrival and service processes so that they can be approximated by Brownian motions, which are the stochastic process analogs of Gaussian random variables. Here, we take a different approach: we will show that heavy-traffic results can be obtained by extending the ideas behind the discrete-time Kingman bound presented in Chapter 3. The analysis of the large-deviations regime in this chapter refines the probability of overflow estimates obtained using the Chernoff bound, also in Chapter 3.

Roughly speaking, heavy traffic refers to a regime in which the mean arrival rate to a queueing system is close to the boundary of the capacity region. We will first use a Lyapunov argument to derive bounds on the moments of the scaled queue length ∈q(t) for the discrete-time G/G/1 queue; where ∈ = μ — λ; μ is the service rate and λ is the arrival rate.

In Chapter 2, we learned about routing algorithms that determine the sequence of links a packet should traverse to get to its destination. But we did not explain how a router actually moves a packet from one link to another. To understand this process, let us first look at the architecture of a router. Generally speaking, a router has four major components: the input and the output ports, which are interfaces connecting the router to input and output links, respectively, a switch fabric, and a routing processor, as shown in Figure 4.1. The routing processor maintains the routing table and makes routing decisions. The switch fabric is the component that moves packets from one link to another link. In this chapter, we will assume that all packets are of equal size. In reality, packets in the Internet have widely variable sizes. In the switch fabric, packets are divided into equal-sized cells and reassembled at the output, hence our assumption holds.

Earlier, we implicitly assumed that this switch fabric operates infinitely fast, so packets are moved from input ports to output ports immediately. This allowed us to focus on the buffers at output ports. So, all our discussions so far on buffer overflow probabilities are for output queues since an output buffer is the place where packets “enter” a link. For example, WFQ, introduced in Chapter 3, may be implemented at output port buffers to provide isolation among flows. However, in reality, the switch fabric does not really operate at infinite speed.