The Energy Efficient Mobile Cloud Computing

Published Date: 02 Nov 2017

Abstract— According to the demand with the rapid increment of Mobile Cloud Computing (MCC), huge amount of application and services appear on mobile devices. As we know, limited battery power is considered a serious problem for mobile devices. One of the most popular applications is location based applications (LBA) which is energy hungry application. GPS (Global Positioning System) which is a very useful application and it is also notorious for its hunger for power. An ample amount of research is going on for energy–efficient locating and sensing system. This paper gives a practical survey on several low-power designed LBA and an energy-efficient mechanism for Green Cloud Computing. We illustrate the characteristics of some practical measurements with few popular cell phones which demonstrate general energy savings comparison.

Index Terms- Mobile cloud computing, Green cloud computing, Location based services, Energy-efficiency, Off-loading.

Introduction

Mobile phones become a vital part of our present life. It makes our life more convenient and effective in terms of time and place. Recent progress in Mobile Computing becomes a significant trend in the development of IT technology. Mobile Cloud Computing is an integration of Cloud Computing and mobile networks which brings novel types of services and facilities for users. It has features like lowering operating cost, low power requirement and convenient of use. With the characteristics of user; energy consumption, application management, uses these factors vary. Location based services (LBS) are very useful and popular among all users. But it’s a very energy hungry application. As the battery manufacturing industry improves their product slowly (battery capacity grows by only 5% annually [1]) and demand of computing and storage capability is increasing so fast, efficient service within limited battery consumption becomes a major demand. A lot of research is going on about these matters.

The Location based services (LBS) make use of the geographical position of mobile device which have the advantages of both user mobility and cloud resources. By utilizing GPS, these services achieve user’s current position and provide different location-related services. Research shows that user can use GPS for 9 hours on smartphones continuously, which becomes a major issue for mobile locating and sensing.

Apart from that, computing to communication ratio is also a critical factor for the decision between local processing and computation offloading. The interchange is strongly depending on the energy efficiency of wireless communication and local processing. Moreover, the traffic pattern at the concurrent time is also a major factor for energy issue. It will consume more energy if the user sends sequence of data packets instead of sending the same data in a single burst.

In this paper, we represent a comprehensive survey about energy-efficiency of MCC which will be regarded as Green Cloud Computing. We took several mobile phones to see the battery performance using GPS, WiFi and GSM network.. To find the proper way of energy-efficiency, we examined several sensing based technology and Off-loading effect.

Overview of Mobile Cloud Computing

Mobile Cloud Computing plays a significant role in current cellular technology. It becomes a revolution for mobile users which helps to achieve phenomenal experience of a variety of mobile applications at low cost. It also attracts the entrepreneur as a demanding and profitable business criterion that improves the availability of mobile applications significantly. Cloud computing facilitates the tasks when data are stored on the internet rather than on a particular device, which provide on-demand access [4]. The applications in the mobile device which is connected to the internet are mainly run on a remote server.

Several methods of energy-efficient locating sensing are available. Maintaining position accuracy is still a challenge which is the motivation of Dynamic Tracking. Here we have discussed the application related optimizations is some specific categories.

Dynamic Tracking

In dynamic tracking, every time it attempts to minimize the frequency of needed position updates by just sampling positions (GPS), when the estimated uncertainty in position exceeds the accuracy threshold. It takes into account a constant positioning accuracy and delay, target speed and acceleration to detect if the target is moving or not. It assumes that the parameters mentioned are constant.

EnTracked, RAPS and EnTrackedT show a similar system structure, while dissimilar in some technologic details. These three works represent the most typical instances of dynamic tracking, and will be discussed below.

EnTracked [14] generally uses an accelerometer to detect any movement. It uses an energy model to dynamically compute parameters such as the delays and consumption which can describe the power consumption of a real phone with a much higher precision. Speed is approximated using the speed and accuracy provided by the GPS module. The error limit (accuracy threshold) is previously given to EnTracked. After this, the point at which to power features (mainly GPS and radio) on and off is calculated from the parameters estimated above and the device model.

In EnTrackedT, the idea of trajectory tracking corresponding to position tracking in EnTracked is being used. EnTracekedT focuses on a current position. At first, EnTrackedT adopts a Heading-Aware Strategy, which employs the compass as a turn point sensor and significantly reduces power consumption of trajectory tracking. EnTrackedT measures the accumulated distance traveled orthogonal to the initial position given by the compass, and compares this to the prescribed trajectory error threshold. We can see clearly that intervals between GPS usage can be much larger than in EnTracked, as seen in Fig.1. Secondly, EnTrackedT uses adaptive duty cycling strategies for the accelerometer and compass sensors, which make the system more powerful. Thirdly, EnTrackedT uses a speed threshold based strategy together with an accelerometer based strategy for movement detection. This strategy enables the system to handle several transportation modes e.g., walking, running, biking or commuting by a car. Fourthly, it explored algorithms of a modified but simple motion trajectory to reduce data size and communication costs caused by sending motion information.

Fig. 1: Heading deviations will increase the orthogonal distance beyond the threshold and force the GPS position to be updated.[18]

The error percentage of the EnTrackedT system is comparatively high when the requested error threshold is small while the power consumption is much lower at the same time. Although EnTrackedT proclaims to have joint trajectory and position tracking, it seems to work better for trajectory based applications.

RAPS [20] is based on the observation that GPS is generally less accurate in urban areas. It introduces the concept of activity ratio, which is the fraction of time that the user is in motion between two position updates. It uses an accelerometer to detect movement while measuring the activity ratio at the same time. It then uses this activity ratio along with the history of velocity information to estimate the current velocity of the user. RAPS duty-cycles the accelerometer carefully, using a duty-cycling parameter deduced empirically. A significant portion of the energy savings of RAPS come from avoiding GPS activation when it is likely to be unavailable to use celltower-RSS (the received signal strength) blacklisting. It records the current celltower ID and RSS information and associates with the success or failure of GPS. Additionally, RAPS uses Bluetooth to share the newly updated position information to save more energy. RAPS uses a combination of spatiotemporal location history, user activity, and celltower-RSS blacklisting and it also proposes sharing position readings among nearby devices. However, it has limitations as well. First, RAPS is mainly designed for pedestrians in urban areas. Second, the user space-time history and the celltower-RSS blacklist must be populated for RAPS to work properly. Third, its velocity estimation based on activity ratio can be misled by handset activity not related to human motion. Fourth, accelerometers on smartphones may need a onetime, per-device calibration of the offset and scaling before running RAPS.

Alternative positioning technologies

Global Positioning System is used very often for dynamic tracking. As periodic or adaptive duty-cycling of GPS may not achieve ambient energy savings under all conditions, several works have explored schemes which would rarely use GPS for positioning. These strategies are based on the spatio-temporal consistency in user mobility, or the large population statistics on routes in an area. These strategies are also integrated with GPS-assisted training.

CAPS [12] uses a Cell-ID Aided Positioning System based on the consistency of traveled routes and consistent cell-ID transition points. It saves the history of cell-ID and GPS position sequences, and then senses the cell-ID sequences to estimate the current position using a cell-ID sequence matching technique. According to the observation, for mobile users with consistent routes, the cell-ID transition point for each user can often uniquely represent the current user position. It is designed for highly mobile users who travel long distances in a predictable fashion. It will be unable to work in some cases where GPS is not available such as indoors and the size of the historical database may be very large if the user travels much. Also, it is evaluated only in urban areas where cell-tower density is high. CAPS does not make use of the underlying geography like Rural area or in plain land or forest as our observation.

EnLoc [2] also explored how to make use of the spatio-temporal consistency in user mobility. When exploiting habitual mobility, EnLoc uses the logical mobility tree (LMT) to record the person’s actual mobility paths showed in Figure 2. The vertices of the LMT are also offered to as uncertainty points. The basic idea is to sample the activity at a few uncertainty points, and EnLoc assumes the rest.

8:00

School

Home

8:10 AM

8:20

6:10

6:00

Gym

7:10

7:00

Wal-Mart

Library

Fig.2: personal mobility profile: A spatio-temporal LMT

The procedure mentioned above highly relies on, as well as limits to the spatio-temporal consistency in user mobility. It cannot tackle users’ deviation from habits. So EnLoc further exploits mobility of large populations as a potential indicator of the individual’s mobility.

EnLoc hypothesizes that a "probability map" can be generated for a given area from the statistical behavior of large populations. Then an individual’s mobility in that area can be predicted. For example, considering a person approaching a traffic intersection of street A: since the person has never visited this street, it is hard to predict how he/she will behave at the imminent intersection. However, if most people are used to take a left turn to Street B, the person’s movement can be inferred accordingly. Otherwise, it has to calculate a lot.

Multiple LBA Management

More than one location based application may run on a single smartphone at a time, the asynchronous calls of GPS from different LBAs unnecessarily lead to higher energy cost. LBAs [11] presents a design principle called Sensing Piggybacking (SP) to overcome this shortcoming. It proposes a middleware to manage multiple LBAs to avoid unnecessary GPS invoking events.

SP listens to the sensing requests of LBAs and forces the incoming registration request to synchronize with existing location-sensing registrations. LBAs uses a triple (G1,T1,D1) to describe the location sensing requirement of the joining LBA, where G1 is the granularity of sensing (e.g.,fine (or GPS) and coarse (orNet)), T1 is the minimum time interval and D1 is the minimum distance interval for location updating. It uses (G2, T2,D2) to denote the finest existing GPS registration, where T2 and D2 are the finest sensing intervals. Similarly, it uses (G3 T3,D3) to denote the finest Net registration. The incoming triple is compared with the existing registration, and SP makes decision whether to register a new request or simply use the current one according to the granularity and interval requirements. It can re-use the existing sensing registrations which eliminate some location-sensing invocations as well as save energy.

Since more than one LBA may be running on one smartphone at the same time, a middleware of multiple LBA management is essential for energy-efficient sensing. This middleware should be reconfigured when incorporated with other energy-efficient mechanisms, just like SP is used with other principles in LBAs.

Trajectory Simplification

Trajectory simplification has been proposed as a way to reduce data size and communication costs caused by sending motion information. It is used for applications which need trajectory information instead of a single position. The main idea of trajectory simplification is to use a smaller subset of obtained a vintage point which is minimal in size while still reflecting the overall motion information. In EnTracked, trajectory simplification is viewed as a special case of line simplification (which has been thoroughly discussed in the computational geometry community).

The main consideration of trajectory simplification is the trade-off between computation cost and simplification. EnTrackedT designed several algorithms and made comparison. The power consumptions of several algorithms are measured to choose the suitable one and it may be relevant mobile systems.

Offloading

Sending computation to another computer is an old trend. What distinguishes cloud computing from existing model is the adoption of virtualization. Instead of service providers managing programs running on servers, virtualization allows cloud vendors to run arbitrary applications from different customers on virtual machines. Cloud vendors thus provide computing cycles and users use these cycles which reduce the amounts of computation on mobile system and save energy. Thus Cloud computing can save energy for MCC through Offloading.

Experimental Analysis

We examined performance of several dynamic tracking and Alternative Positioning technologies to investigate which one is energy efficient. We made a comparison between RAPS and ENLOC to investigate energy consumption and result is in Table: 1 below.

Mobile Model

RAPS Energy Consumption (In Percent)

ENLOC Energy Consumption (In Percent)

Samsung Exhibit II

21%

5%

Samsung Galaxy S II

20%

3%

Samsung Galaxy S III

19%

4%

Google Nexus 4

20%

3.5%

HTC Evo

21%

5%

HTC MyTouch 4G

22%

5 %

Table 1: Percentage energy consumption of Several Mobile Model

Fig.3: Graphical representation of Table 1.

We can see ENLOC is more efficient than RAPS in some cases. But RAPS has upper hand in accuracy. Now we calculate how much energy we can save by Offloading.

Let’s assume the computation requires C instructions. If M and S be the speeds, in instructions per second, of the cloud server and the mobile system, respectively. The same task thus takes C/S seconds on the server and C/M seconds on the mobile system. If the server and mobile system exchange D bytes of data and B is the network bandwidth, it takes D/B seconds to transmit and receive data. The mobile system consumes, in watts, Pc for computing, Pi while being idle, and Ptr for sending and receiving data

If the mobile system performs the computation, the energy consumption is Pc× (C/M). If the server performs the computation, the energy consumption is [Pi× (C/S)] + [Ptr× (D/B)]. The amount of energy saved is

Pc × () – Pi × ( ) - Ptr × (1)

Suppose the server is F times faster—that is, S= F× M. We can rewrite the formula as

× ( Pc - ) - Ptr × (2)

Energy is saved when this formula produces a positive number. The formula is positive if D/B is sufficiently small compared with C/M and F is sufficiently large. The values of M, Pi, Pc, and Pth are parameters specific to the mobile system. If we use a four-core server, with a clock speed of 3.2 GHz, the server speedup F may be given by (S/M) ≈[(3.2 × 1,024 × 4 × X)/400], where X is the speedup due to additional memory, more aggressive pipelining, and so forth. If we assume X= 5, we obtain the value of F≈ 160.

The value of F can increase even more with cloud computing if the application is parallelizable, since we can offload computation to multiple servers. If we assume that F= 160, Equation 2 becomes

( 0.9 - ) - 1.3 × = (0.0025 × C) - 1.3 × (3)

We equate Equation 3 to zero and obtain for offloading to break even,

Bo= 577.77 ×

Where, Bo is the minimum bandwidth required for offloading to save energy, determined by the ratio of (D/C). If (D/C) is low, then offloading can save energy. Thus, offloading is beneficial when large amounts of computation Care needed with relatively small amounts of communication D.

IX. Conclusion

From our experiment and calculation we see that we have to make some tradeoff to achieve Green Cloud Computing. For the case of accuracy we should use RAPS and for daily work we should switch to CAPS for energy efficiency. For this case, we can construct an algorithm which will switch between ENLOC and RAPS. This can be done as a future work.

We also see Offloading will be a nice solution for Green Cloud Computing. For this case, we can use smart grid in Data Center which will be more energy efficient.

Acknowledgement

The authors would like to thank Dr. Naima Kaabouch for her continuous support and lively discussions about cloud computing. Her book "Sustainable ICTs and Management Systems for Green Computing" truly helps us to understand the fundamental topics. The views expressed in this paper are those of the authors and do not necessarily reflect the views within partners.