TY - JOUR
AU - Mukherjee, Nandini
AB - Abstract The use of cloud computing and mobile devices is increasing in healthcare service delivery primarily because of the huge storage capacity of cloud, the heterogeneous structure of health data and the user-friendly interfaces on mobile devices. We propose a healthcare delivery scheme where a large knowledge base is stored in the cloud and user responses from mobile devices are input to this knowledge base to reach a preliminary diagnosis of diseases based on patients’ symptoms. However, instead of sending every response to the cloud and getting data from cloud server, it may often be desirable to prune a portion of the knowledge base that is stored in a graph form and download in to the mobile devices. Downloading data from cloud depends on the storage, battery power, processor of a mobile device, wireless network bandwidth and cloud processor capacity. In this paper, we focus on developing mathematical expressions involving the above mentioned criteria and show how these parameters are dependent on each other. The expressions built in this paper can be used in real-life scenarios to take decisions regarding the amount of data to be pruned in order to save energy as well as time. 1 Introduction Technologies that mobile devices, wireless network and cloud computing together have made enormous progress in the past decade and have provided a ubiquitous computing and communication environment for real-world applications. We propose a scheme for mobile-assisted remote healthcare service delivery in [1, 2]. The scheme has been proposed to assist preliminary diagnosis for patients in a remote healthcare delivery framework through mobile and cloud collaboration. The scheme has been presented in Fig. 1. Usually, a health assistant (HA) helps the patients undergo the process of preliminary diagnosis of the illness. A knowledge-based Disease-Symptom graph data model has been built following the guidelines of medical practitioners. The graph data model contains the diseases, symptoms and other necessary information or signs of symptoms that help to classify the appropriate symptom and to identify the disease through graph traversal techniques. The traversal takes place through a series of key questions. A sample knowledge-based disease-symptom data model is shown in Fig. 2. The knowledge base is stored in the cloud. Initially, when the patient visits HA, HA enters multiple symptoms, related to the patient’s illness through the mobile device. Then, a query formulation module extracts a subgraph from the knowledge base based on the symptoms. The subgraph is used to display a set of questions on the HA’s mobile phone and the HA seeks answers to those questions from the patient. Based on the responses and the initial symptoms, the preliminary diagnosis decision-making module identifies the disease. Figure 1. Open in new tabDownload slide Scheme of mobile-assisted remote healthcare. Figure 1. Open in new tabDownload slide Scheme of mobile-assisted remote healthcare. The Disease-Symptom knowledge graph, which assists the care providers to make preliminary diagnosis of diseases, is large and complex and needs to be stored in a large storage, such as in the cloud. Also, it is clear that the diagnosis process requires traversal of the Disease-Symptom graph based on the patients’ responses to few predefined queries. However, repeated queries and responses while traversing the graph may lead to large number of packet transmissions between cloud and mobile devices causing unnecessary delay. Our remote healthcare delivery framework is being deployed in rural areas. Therefore, the HAs need to work under constraints like low bandwidth and battery-powered mobile devices. Moreover, due to resource constraints of the mobile devices, it is also not possible to load the entire graph onto the mobile devices. It has been therefore proposed that whenever a patient visits the HA with some symptoms that are then entered into the system, a portion of the knowledge graph rooted at the input symptoms may be pruned and downloaded on the mobile devices [3]. Hence, the graph can be traversed only locally based on the queries and responses from the patients. An android app that implements the pruning strategy has been demonstrated in a previous research work [4]. The app helps to understand how the strategy can be used for rapid diagnosis of diseases in a mobile-assisted remote healthcare delivery scheme. Though, pruning seems to be a good solution in many situations, it may not always be profitable or even practical. If a large portion of the knowledge graph needs to be pruned for a set of symptoms, that may not be possible because of the limited resources of mobile devices, like storage and processing power and the bandwidth of the wireless network. Therefore, there are two options: (i) to prune the directed acyclic subgraph rooted at multiple symptoms and perform the processing locally or (ii) to send the symptoms and intermediate questions repeatedly in the cloud and execute the entire process in cloud only. A hybrid of these two options can also be considered. Hence, a decision should be taken regarding whether to process queries (traversing the knowledge graph) in the cloud or download the subgraph to the mobile device for local processing. The decision must be taken dynamically based on the parameters affecting the processing time. In the mobile devices, battery power, processing capability and mobile storage are the main factors that affect the processing time. On the other hand, bandwidth of the wireless communication and cloud processing capability are other key challenges that affect the query processing time in the cloud. The scheme that we have proposed for mobile-based healthcare service delivery [1, 2] is briefly shown in Fig. 3. In this scheme, a Disease-Symptom graph database is stored on the cloud side and a set of Symptoms are input (⁠|$query$|⁠) through the mobile device. After reaching the cloud, the query is processed (symptoms are matched with the of nodes of the Disease-Symptom graph) and a subgraph is pruned and downloaded to the mobile device. Based on the constraints shown in Fig. 3, this paper focuses on finding the answer to the question that while performing the mobile-assisted remote healthcare service delivery, how much data is to be pruned and downloaded from cloud to mobile device such that the efficiency of query processing is increased and cost of communication is optimized. In this paper, mathematical expressions are developed to analyze the scenarios maintaining all the constraints on the cloud side, as well as on the mobile side. Figure 2. Open in new tabDownload slide Snapshot of real Disease-Symptom database. Figure 2. Open in new tabDownload slide Snapshot of real Disease-Symptom database. Figure 3. Open in new tabDownload slide Key factors of mobile-cloud communication in mobile-assisted remote healthcare. Figure 3. Open in new tabDownload slide Key factors of mobile-cloud communication in mobile-assisted remote healthcare. The remaining part of the paper is organized as follows. In Section 2, we review the contemporary related research works to find how the constraints of mobile devices, cloud data centers and transmission medium affect the communication or data transfer. In Section 3, mathematical expressions are built up to form the basis of our work to show how our proposed scheme is benefited with the use of pruning technique. Next, in Section 4, we analyze the conditions for determining how the energy consumption on mobile devices or data transfer are affected by bandwidth, number of data transmissions, size of the pruned data, as well as transmission power. The paper is concluded in Section 5. 2 Related Work In the recent past, many emerging real-life applications have been built using a collaboration of mobile and cloud technologies. Due to the resource constraint of mobile devices, mobile energy consumption is one of the main issues for this type of real applications. The data transmission between mobile and cloud is also another important issue that affects the mobile battery power. In the previous works [3, 5], pruning is used to download a subgraph from a large, complex knowledge graph from the cloud. Minimizing mobile energy consumption and time delay are the main two purposes of pruning. There are lots of research works that focus on issues related to energy consumption on mobiles and transmission delay in mobile cloud applications. In [28], the authors have conducted a literature survey to investigate different techniques for reducing the power consumption in smartphones. They discussed the components responsible for power consumption in smartphones, such as processors, memory, display, wi-fi and 3G-4G transceivers, GPS, bluetooth and camera. Also, the authors have presented the amount of power saving in smartphones. In another recent survey paper [29], authors have analyzed the causes of power consumption in smartphones in detail and also suggested different measures to minimize the consumption related to several factors. These papers include literature reviews on assessment and estimation of power consumption in smartphones, energy saving techniques and mitigating the hazardous issues of smartphone batteries. The contemporary research works are divided into three core categories: energy measurements and power models for reducing mobile energy consumption; adopting various techniques for reducing the mobile energy consumption and increasing performances of the mobile devices; addressing various constraints of mobile as well as communication medium. Figure 4. Open in new tabDownload slide Literature review on mobile energy and time saving. Figure 4. Open in new tabDownload slide Literature review on mobile energy and time saving. Energy measurements and power models. A number of power models have been evaluated to understand the energy consumption by different hardware components. In [5], the authors have shown that mobile energy consumption actually reduces the energy consumption by different hardware components. In this paper, openmoke NCE Freerunner phone has been used for the purpose of experiment and the breakdown of power consumption has been done by external high resolution meter. The authors have also shown that the energy is consumed mostly by GSM module and the display unit. In another paper [6], the authors have shown that energy consumption depends on three services, i.e. text messaging and voice and data communication. Also, here, the authors have concluded that the energy consumption is higher in 3G network than in 2G network for text messaging and voice communication. In another recent work [30], the authors have explored the energy estimation and power models in literature reviews. They classify the existing energy estimation and modeling schemes of smartphone applications into two categories—code analysis and mobile components power model-based estimation. They further classify code analysis based modeling and estimation techniques into simulation-based and profiling-based categories. Moreover, the authors highlight the commonalities and differences in the existing energy estimation and modeling estimation schemes based on a few parameters. In [31], the authors have proposed a service-specific end-to-end energy-efficiency modeling to demonstrate how measurements can be conducted and used in service-specific end-to-end energy consumption assessments. Their main finding is that web browsing, instant messaging and wireless network are the main energy consuming services. The authors further concluded that mobile energy can be reduced in popular and emerging applications by using small-cell offloading and mobile edge caching. Some other researchers have also analyzed mobile energy consumption using different power models and energy measurements in their research papers [7–12]. Techniques for reducing mobile energy. Due to resource constraints of mobile devices, part of the compute-intensive applications are offloaded from mobile devices onto cloud. This technique is called offloading. There are many research works [14, 15, 18, 19] that have been carried out to reduce the mobile energy consumption by offloading compute-intensive programs to cloud. The objectives of offloading schemes can be divided into three categories, i.e. energy saving, time saving and a combination of both. A few research papers [13, 17, 20] focus on reducing mobile energy only. For these schemes, after energy profiling each part of an application, a cost graph is generated. The nodes and edges of the cost graph are represented as functions and data that are to be transmitted to the cloud. It can be partitioned with the use of max-flow/min-cut algorithm to transfer the heavy task to cloud from mobile devices. In [25], the authors have proposed a systematic approach, EFFORT, for offloading computation in the cloud to solve energy consumption problem of smartphones. First, they have presented how battery power consumption affects the performance of smartphone applications, storage and computation. Later, they have shown that with offloading, energy consumption in smartphones is reduced by 19% and data usage reduced by 16%. On the other hand, reducing the execution time by offloading the data or part of program code to remote server or cloud has been dealt with in [12, 16]. Techniques for both energy and time saving are proposed by several research works [21, 22] by optimizing two objectives simultaneously. Analyzing different constraints of mobile devices, as well as wireless medium. In many research papers, the authors have shown how energy consumption on mobile devices and data transfer are affected by various properties or constraints [23, 24, 26]. In these research papers, the authors have also proposed solutions to prolong battery life time of mobile devices. In particular, constraints of mobile devices, cloud, as well as that of wireless mediums have been discussed in [25, 27] along with the issues related to energy efficiency of mobile devices. In [13], the authors have discussed that data transmission over wireless channels incurs power consumption, thereby posing challenge for implementation of mobile cloud applications. The research work in [15] has analyzed the energy consumption for a computational task on mobile devices and compared the same on the server in a cloud. In this paper, the authors have highlighted various mobile constraints, as well as constraints on cloud and characteristics of a wireless medium. In addition to the above research papers, in [32], the authors have measured the energy consumption of a cell phone for WBAN applications while transmitting and encrypting data. The authors recommended an optimal strategy based on file size, data communication network and encryption algorithm for reducing energy consumption and improving life time of WBAN. For experimental evaluation, an android app has been created to send data by uploading it to the local server and encrypting the data using existing AES algorithm. For energy calculation, power tutor has been used. They concluded that bigger-sized files were transmitted less and consumed lesser energy. In another research paper, [33], the authors have highlighted the factors related to mobile power consumption like wireless connection, fluctuation of communication bandwidth and user mobility. The authors have concluded that more energy is consumed due to bad connectivity. They have also presented a eTime data transmission strategy between cloud and mobile devices based on Lyapunov optimization. The authors claimed that eTime can be applied to various popular applications while achieving 20%–35% energy saving. An experimental study of the energy consumption behavior of representative data mining algorithms running on mobile devices has been presented in [34]. In this paper, the authors reveal that by appropriate tuning of a few parameters like data set size, number of attributes, size of produced results, etc., data mining algorithms can be efficiently executed on mobile devices by saving energy and increasing the lifetime of the devices. Also, a machine learning approach has been proposed in this paper to predict energy consumption of mobile data intensive algorithms. However, in the above research papers, no analysis has been done to support the decision regarding how much data is to be transferred from cloud to mobile devices, so that efficiency of query processing is improved. In this paper, our endeavour is to conceptualize how the energy consumption on mobile devices and data transfer from cloud to mobile devices depend on various constraints of mobile devices, wireless channel, instruction sets and cloud properties. The processing speed on mobile devices and on cloud and data transmission between mobile devices and cloud are taken into account in this regard. 3 Analysis based on amount of pruned data In our previous work [1], we described a framework of mobile-assisted remote healthcare service delivery. A large complex Disease-Symptom graph is stored as a knowledge base in graph database. In a rural or remote area, where doctor is not available, an HA helps the patient to diagnose the disease by asking a series of pre-determined questions that are related to the sign of symptoms of the patient. The symptoms, sign of symptoms and the diseases form various nodes of the Disease-Symptom graph and patient’s responses to the questions lead to different traversals through the graph, finally leading to appropriate diagnosis for a particular patient. According to the above scheme, while diagnosing a particular patient, communication between mobile device and cloud can be possible in two different ways. Firstly, the Disease-Symptom graph remains in the cloud and based on the patient’s symptoms, |$S_{1}$|⁠, |$S_{2}$|⁠...|$S_{n}$| individually, the respective sign of the symptoms (intermediate nodes from the graph) are transferred as results. In Fig. 5, it is shown for only one symptom, |$S_{1}$|⁠. HA or patient sends the symptoms through a mobile device. After acceptance of symptom, results (⁠|$A_{1}$|⁠) are sent from the cloud as sign of the symptoms. These results are presented to the user (HA/patient) as questions and the user response (⁠|$R_{1}$|⁠) is sent to the cloud. The response helps to traverse further down through the Disease-Symptom graph and the next set of sign of the symptoms (⁠|$A_{2}$|⁠) is sent to the mobile devices and the response from the HA/patient (⁠|$R_{2}$|⁠) is sent back again. The same procedure is continued until a diagnosis related to a particular disease (a leaf node of the Disease-Symptom graph) is reached. This process is implemented for all symptoms of the patient. During this communication, a lot of packets are transmitted between the mobile device and cloud as shown in Fig. 5. Secondly, only symptoms are asked through a query and the set of main symptoms are matched with the symptoms in the knowledge base. Based on this input, a subgraph is pruned and downloaded to the mobile device from cloud. After that, the questions or sign-of-the-symptoms are asked locally (not to the cloud) for the diagnosis of the patient. The second case is shown in Fig. 6 in which only a single |$query$| (as set of symptoms) is forwarded to the cloud from a mobile device and |$answer$| or a subgraph is pruned from the cloud to the mobile device. Figure 5. Open in new tabDownload slide First way to query for mobile-assisted remote healthcare. Figure 5. Open in new tabDownload slide First way to query for mobile-assisted remote healthcare. Figure 6. Open in new tabDownload slide Second way to query for mobile-assisted remote healthcare. Figure 6. Open in new tabDownload slide Second way to query for mobile-assisted remote healthcare. In order to explain clearly the above-mentioned cases, an example is given here with the Disease-Symptom graph database shown in Fig. 7. Symptoms and sign of symptoms are sent one by one as shown in Fig. 8. Here, |$S_{1}$| is sent as |$query$| from mobile to cloud and possible sets of sign of symptoms are returned as results that are represented to the patient as questions. Then, by giving answers to the questions (selecting appropriate sign of the symptom) from a mobile device, the next response is sent to the cloud. In this way, finally a list of possible |$diseases$| is returned. In this process, lots of data are transferred from mobile device to cloud and from cloud to mobile device. The transferred data size can be expressed as |$\pi _{1}$| for the first case. The transferred data from cloud to mobile device must be less than the memory size (⁠|$M$|⁠) of the mobile device. Figure 7. Open in new tabDownload slide An example of Disease-Symptom graph database. Figure 7. Open in new tabDownload slide An example of Disease-Symptom graph database. Figure 8. Open in new tabDownload slide First procedure to diagnosis of a patient. Figure 8. Open in new tabDownload slide First procedure to diagnosis of a patient. On the other hand, in the second case, set of |$symptoms$| are sent as a |$query$| as shown in Fig. 9 that are matched as a whole and the subgraph (⁠|$\pi _{2}$|⁠) is returned to the mobile device. The subgraph is shown against the input of |$S_{1}$| and |$S_{2}$| in Fig. 10. In this case, data size(⁠|$\pi _{2}$|⁠) is certainly larger than the previous data size(⁠|$\pi _{1}$|⁠). This is because in the second case the subgraph is returned, while in the other case, only answers are returned back to the mobile devices. Figure 9. Open in new tabDownload slide Queries for the pruning concept. Figure 9. Open in new tabDownload slide Queries for the pruning concept. Figure 10. Open in new tabDownload slide Pruned subgraph for the input of |$S_{1}$| and |$S_{2}$| of Disease-Symptom graph database. Figure 10. Open in new tabDownload slide Pruned subgraph for the input of |$S_{1}$| and |$S_{2}$| of Disease-Symptom graph database. Table 1 depicts the commonly used symbols in this paper. Table 1. Symbols and descriptions. Symbols Descriptions |$\gamma $| Instruction length of queries |$\mu $| Mobile processor speed |$S_{i}$| Symptoms of Disease-Symptom graph |$\lambda $| Cloud or server processor speed |$\psi $| How much times speed cloud than mobile processor |$\pi $| Prune data for each query |$\beta $| Wireless network bandwidth |$\rho _{m}$| Mobile processing or query computation, energy |$\rho _{t}$| Transmission power |$\rho _{i}$| Energy consumption for mobile idle state |$\xi $| Total energy consumption |$q_{n}$| |$n$| number of query |$t$| Time Symbols Descriptions |$\gamma $| Instruction length of queries |$\mu $| Mobile processor speed |$S_{i}$| Symptoms of Disease-Symptom graph |$\lambda $| Cloud or server processor speed |$\psi $| How much times speed cloud than mobile processor |$\pi $| Prune data for each query |$\beta $| Wireless network bandwidth |$\rho _{m}$| Mobile processing or query computation, energy |$\rho _{t}$| Transmission power |$\rho _{i}$| Energy consumption for mobile idle state |$\xi $| Total energy consumption |$q_{n}$| |$n$| number of query |$t$| Time Open in new tab Table 1. Symbols and descriptions. Symbols Descriptions |$\gamma $| Instruction length of queries |$\mu $| Mobile processor speed |$S_{i}$| Symptoms of Disease-Symptom graph |$\lambda $| Cloud or server processor speed |$\psi $| How much times speed cloud than mobile processor |$\pi $| Prune data for each query |$\beta $| Wireless network bandwidth |$\rho _{m}$| Mobile processing or query computation, energy |$\rho _{t}$| Transmission power |$\rho _{i}$| Energy consumption for mobile idle state |$\xi $| Total energy consumption |$q_{n}$| |$n$| number of query |$t$| Time Symbols Descriptions |$\gamma $| Instruction length of queries |$\mu $| Mobile processor speed |$S_{i}$| Symptoms of Disease-Symptom graph |$\lambda $| Cloud or server processor speed |$\psi $| How much times speed cloud than mobile processor |$\pi $| Prune data for each query |$\beta $| Wireless network bandwidth |$\rho _{m}$| Mobile processing or query computation, energy |$\rho _{t}$| Transmission power |$\rho _{i}$| Energy consumption for mobile idle state |$\xi $| Total energy consumption |$q_{n}$| |$n$| number of query |$t$| Time Open in new tab 3.1 Energy consumption by the mobile devices It has already been mentioned that the downloading of pruned subgraph is totally dependent on resources available on the mobile devices, characteristics of wireless communication and processors on the cloud or server. Resources of the mobile devices are as follows: |$\bullet $| mobile battery power or energy consumption, |$\bullet $| mobile storage, |$\bullet $| mobile device processing capability (or processor ability) and |$\bullet $| mobile transmission power. When mobile devices communicate with cloud or server at data center, the required time can be divided into three stages as follows: |$\bullet $| first step is mobile query processing that happens on the mobile devices; |$\bullet $| second step is sending data to the cloud or receiving data from the cloud by mobile devices; and |$\bullet $| third step is the processing time (or traversal time) that happens in the cloud or server side. Energy consumption by the mobile devices increases when mobile devices perform data processing, and transmit data even when mobile devices are in the idle state. Let |$t_{1}$| be the processing time on cloud, |$t_{2}$| be the transmission time (assuming the time for both sending query and receiving data) and |$t_{3}$| be the time required for mobile processing or computation time on the mobile device. When a query is processed on the cloud in |$t_{1}$| time (on the basis of its processing ability), the mobile device remains in idle state at that time. In order to distinguish between the above two, we define the following. Mobile processing. When query is sent to the mobile device, then the processing involves interpretation, transformation of the query to location specific query, use of local cache, etc. All these processing activities all together is referred to as mobile processing. Cloud processing. Cloud processing implies the execution of the query in cloud environment. Figure 11 depicts the processing on the cloud and on the mobile device. The mobile device consumes energy during the entire process (total time), i.e. for |$t$|= |$t_{1}$|+|$t_{2}$|+|$t_{3}$|⁠. Figure 11. Open in new tabDownload slide Measures of energy and time for mobile devices. Figure 11. Open in new tabDownload slide Measures of energy and time for mobile devices. During transmission of data, mobile battery power is one of the major criteria. Suppose a query with |$\gamma $| instructions is processed on a mobile device. Same query with |$\gamma $| instructions can be executed in cloud or server in the data center. Thus, the query processing time |$t_{3}$| on a mobile device is represented as |$t_{3}=\gamma /\mu $|⁠, where |$\mu $| is the processing speed of the mobile processor. When the same query is processed in cloud or server, the required time is |$t_{1}=\gamma /\lambda $| where |$\lambda $| is the speed of the cloud or server processor. Processor speed on server or cloud is much faster than the processor speed of the mobile devices. Let the processing speed on cloud be |$\psi $| times larger than that on the mobile devices. Let |$\pi $| bytes be the eligible data that are transmitted from cloud or server to mobile devices or vice versa. If |$\beta $| is the network bandwidth of the wireless network, then |$\pi /\beta $| is the time needed to transmit or receive data from or to the mobile devices. Hence, the total energy spent is the sum of the energy spent for mobile processing, mobile device to cloud transmission and the reverse and energy consumption during mobile idle state. Hence, total energy consumption = energy consumed for mobile processing + energy consumed for mobile transmission + energy consumed during the idle state. If the per unit time energy consumption by the mobile system for query computation is |$\rho _{m}$|⁠, |$\rho _{t}$| for data transmission and |$\rho _{i}$| for being in idle state, then |$\xi $| is the total energy consumed during |$t_{1}$|⁠, |$t_{2}$| and |$t_{3}$|⁠, which is given by $$\begin{align*} &\xi= \rho_{m}\frac{\gamma}{\mu}+\rho_{i}\frac{\gamma}{\lambda}+\rho_{t}\frac{\pi}{\beta} \end{align*}$$ or $$\begin{align*} & \frac{\pi}{\beta}*{\rho_{t}}= \xi-\frac{\gamma}{\mu}*(\rho_{m}+\frac{\rho_{i}}{\psi}) \end{align*}$$ or $$\begin{align} \pi= \frac{\beta}{\rho_{t}}[\xi-\frac{\gamma}{\mu}(\rho_{m}+\frac{\rho_{i}}{\psi})]. \end{align}$$(1) Now, the above equation can be used to analyze how the amount of downloaded data will be effected by the above-mentioned criteria. It must be noted here that the energy consumption of mobile devices at two different states, that is processing and idle states is fixed, or in other words, depends on the device characteristics. However, transmission power can be increased or decreased for a particular mobile device. 3.2 Different communication scenarios between mobile device and cloud Pruning algorithm [3] helps to prolong the battery life time of low resourceful mobile devices and to reduce the number of packet transmission. It is already explained in [3] that pruning algorithms overcome the challenges of low resourceful mobile devices and communication between mobile device and cloud. In this section, we analyze various constraints and factors affecting the communication between mobile device and cloud. There are four possible scenarios in this context. Suppose |$n$| queries, |$q_{1}$|⁠, |$q_{2}$|⁠..... |$q_{n-1}$|⁠.....|$q_{n}$| are generated to traverse the Disease-Symptom graph. Then, the four scenarios are as follows. |$\bullet $| All |$n$| queries are sent in the form of |$symptom$|s or sign of symptoms to cloud, and for every query, |$\pi _{1}$| amount of data is downloaded to the mobile device, which is shown in Fig. 8. This scenario can be termed as No Pruning. |$\bullet $| Suppose |$k$| queries are sent in the form of |$symptom$|s and sign of symptoms to cloud and |$\pi _{3}$| amount of data is downloaded for every query. Remaining |$n-k$| queries are processed locally in the mobile device. This scenario can be called Partial Pruning. |$\bullet $| Only sets of symptoms as a whole are sent as one single query to cloud and |$\pi _{2}$| amount of data is downloaded to the mobile device as a pruned subgraph. The rest of |$n-1$| queries are given as set of sign of symptoms to the subgraph stored in the mobile devices and processed locally to diagnose a disease. This scenario can be called Pruning. |$\bullet $| In the fourth scenario, all the |$n$| queries are given in the form of symptoms and sign of symptoms to the mobile device. If a mobile device can store the large, complex knowledge base, then there is no need for pruning a subgraph or communication with the cloud. However, it does not make any sense because mobile device cannot accommodate large amount of data. Thus, we consider the first three scenarios here to evaluate the energy consumption of the mobile device. The three scenarios are shown in Fig. 12. Figure 12. Open in new tabDownload slide Three scenarios of communication between cloud and mobile device with n queries. Figure 12. Open in new tabDownload slide Three scenarios of communication between cloud and mobile device with n queries. Scenario 1: no pruning. Let |$n$| queries be communicated to the cloud (assumed response also a query as in Fig. 5), |$\pi _{1}$| amount of data is pruned to the mobile device for individual query. Then, each of the |$q_{1}$|⁠, |$q_{2}$|⁠,...., |$q_{n-1}$|⁠, |$q_{n}$| queries is first transmitted to the cloud and then answer or result reach the mobile device, each query is processed on the mobile device. If |$\xi _{1}$| is the energy consumed for processing |$n$| queries, then $$\begin{align*} &\prime{\xi_{1}}=n*\rho_{m}\frac{\gamma}{\mu},\end{align*}$$ where |$\rho _{m}$| is the energy consumed per unit time at the time of processing, |$\gamma $| is the set of instructions for each query and |$\mu $| is the mobile processing speed. For simplicity, we consider that the queries are homogeneous and contain same number of instructions. If |$\pi _{1}$| amount of data is pruned and communicated through a channel of |$\beta $| bandwidth and |$\rho _{t}$| is the transmission power, then energy consumed |$\prime{\xi _{2}}$| is represented as follows: $$\begin{align*} &\prime{\xi_{2}}=n*\rho_{t}\frac{\pi_{1}}{\beta}.\end{align*}$$ The mobile device remains in idle state after sending each of the queries and the next query cannot given to the mobile device unless and until the |$\pi _{1}$| amount of data does not reach the mobile device. It is shown in Fig. 8. During the idle state (i.e. when cloud processes the queries), the energy consumed is as follows: $$\begin{align*} &\prime{\xi_{3}}=n*\rho_{i}\frac{\gamma}{\lambda}.\end{align*}$$ In the no pruning condition, after receiving the results in mobile devices, user responses are also sent to the cloud through the mobile device and thus the total |$n$| number of queries is sent to the cloud. The entire procedure is shown in Fig. 8. Hence, the total energy consumed in the no pruning condition is $$\begin{align*} &\prime{\xi}=\prime{\xi_{1}}+\prime{\xi_{2}}+\prime{\xi_{3}}\end{align*}$$ or $$\begin{align} \prime{\xi}= n*\rho_{m}\frac{\gamma}{\mu}+n*\rho_{i}\frac{\gamma}{\lambda}+n*\rho_{t}\frac{\pi_{1}}{\beta}. \end{align}$$(2) Scenario 2: partial pruning. Suppose |$k$| queries are sent to cloud and |$(n-k)$| queries are processed in the mobile device. Then, |$q_{1}$|⁠, |$q_{2}$| —- |$q_{k-1}$|⁠, |$q_{k}$| are processed first in mobile device and then transmitted to the cloud, which will be followed by downloading |$\pi _{3}$| amount of data, which will be downloaded for each query. Thus, for processing of |$n$| queries in mobile devices, if energy |$\prime{\prime{\xi _{1}}}$| is consumed, then $$\begin{align*} &\prime{\prime{\xi_{1}}}=k*\rho_{m}\frac{\gamma}{\mu},\end{align*}$$ where |$\rho _{m}$| is the mobile processing power, |$\gamma $| is the instruction set and |$\mu $| is the mobile processing speed. If |$\pi _{1}$| amount data is pruned with |$\beta $| bandwidth and |$\rho _{t}$| transmission power for all |$k$| queries and |$\prime{\prime{\xi _{2}}}$| energy is consumed, then $$\begin{align*} &\prime{\prime{\xi_{2}}}= k*\rho_{t}\frac{\pi_{3}}{\beta}.\end{align*}$$ After pruning, |$\pi _{3}$| amount of data to mobile device, |$(n-k)$| queries are processed in mobile devices, whereas |$k$| queries are processed in the manner similar to the no pruning condition. Thus, the energy consumed by mobile device while being idle is $$\begin{align*} &\prime{\prime{\xi_{3}}}=k*\rho_{i}\frac{\gamma}{\lambda}.\end{align*}$$ Then, energy consumption for |$(n-k)$| queries that are performed on mobile device is $$\begin{align*} &\prime{\prime{\xi_{4}}}=(n-k)\rho_{m}\frac{\gamma}{\lambda}.\end{align*}$$ Thus, the total energy (⁠|$\xi $|⁠) consumed for the scenario 2 is $$\begin{align*} &\prime{\prime{\xi}}=\prime{\prime{\xi_{1}}}+\prime{\prime{\xi_{2}}}+\prime{\prime{\xi_{3}}}+\prime{\prime{\xi_{4}}}\end{align*}$$ $$\begin{align*} &\prime{\prime{\xi}}=k*\rho_{m}\frac{\gamma}{\mu}+(n-k)\rho_{m}\frac{\gamma}{\mu}+k*\rho_{i}\frac{\gamma}{\lambda}+k*\rho_{t}\frac{\pi_{3}}{\beta}\end{align*}$$ $$\begin{align} \prime{\prime{\xi}}=n*\rho_{m}\frac{\gamma}{\mu}+k*\rho_{i}\frac{\gamma}{\lambda}+k*\rho_{t}\frac{\pi_{3}}{\beta}. \end{align}$$(3) The above equation shows how much energy can be consumed in partial pruning condition. Scenario 3: pruning. If |$q_{1}$| query is sent to the cloud and |$\pi _{2}$| amount of data is pruned and downloaded to the mobile device, then the rest |$(n-1)$| queries run on the mobile device on pruned data, and thus, |$n-1$| queries are processed on the mobile device. The energy consumption of mobile device for |$q_{1}$| query is $$\begin{align*} &\prime{\prime{\prime{\xi_{1}}}}=1*\rho_{m}\frac{\gamma}{\mu},\end{align*}$$ and for pruned data transmission, the required energy is $$\begin{equation*}\prime{\prime{\prime{\xi_{2}}}}=\rho_{t}\frac{\pi_{2}}{\beta}.\end{equation*}$$ Also, |$(n-1)$| queries are processed on the mobile device. The energy consumption for processing is $$\begin{align*} &\prime{\prime{\prime{\xi_{3}}}}=(n-1)\rho_{m}\frac{\gamma}{\mu}.\end{align*}$$ After sending query |$q_{1}$| to the cloud, some time is needed for the pruned data to reach the mobile device and the mobile device is in idle state during that time. Hence, $$\begin{align*} &\prime{\prime{\prime{\xi_{4}}}}=\rho_{i}\frac{\gamma}{\lambda}.\end{align*}$$ Therefore, total energy consumption is Scenario 3 is $$\begin{align*} &\prime{\prime{\prime{\xi}}}=1*\rho_{m}\frac{\gamma}{\mu}+(n-1)\rho_{m}\frac{\gamma}{\mu}+\rho_{i}\frac{\gamma}{\lambda}+\rho_{t}\frac{\pi_{2}}{\beta}+\rho_{i}\frac{\gamma}{\lambda}\end{align*}$$ or $$\begin{align} \prime{\prime{\prime{\xi}}}=n*\rho_{m}\frac{\gamma}{\mu}+\rho_{t}\frac{\pi_{2}}{\beta}+\rho_{i}\frac{\gamma}{\lambda}. \end{align}$$(4) Now, it can be concluded from equations |$2$|–|$4$| that the energy consumption on mobile devices depends on how many responses are sent to the cloud server or in other words number of packet transmission between the mobile device and cloud server. In Scenario |$3$|⁠, concept of pruning data is supported and a pruning algorithm is used to prune data from cloud server [3, 5]. In this pruning scenario, mobile device processed |$(n-1)$| number of queries among |$n$| queries. As a result, the number of packet transmission is minimized. But it is noticed that if all queries are sent to the cloud for pruned data (partial and no pruning cases), the number of packet transmission increases and mobile consumes more energy. As a result mobile battery power is wasted. 3.3 Pruning condition in mobile-assisted remote healthcare Pruning algorithm saves both mobile energy and time for our mobile-assisted remote healthcare system. Hence, we need to formulate a condition to decide when pruning will benefit the process by saving energy as well as time. Energy Saving. Already, in the previous section, we present expressions for mobile energy consumption in no pruning and pruning conditions. With pruning, energy is mainly consumed for processing on the mobile device. The energy is less than the energy consumed in no pruning condition due to number of data transmissions. The energy consumption for Scenario 1 with no pruning: $$\begin{align*} &\prime{\xi}= n*\rho_{m}\frac{\gamma}{\mu}+n*\rho_{i}\frac{\gamma}{\lambda}+n*\rho_{t}\frac{\pi_{1}}{\beta},\end{align*}$$ and the energy consumption for Scenario 3 with pruning: $$\begin{equation*}\prime{\prime\prime{\xi}}=n*\rho_{m}\frac{\gamma}{\mu}+\rho_{t}\frac{\pi_{2}}{\beta} +\rho_{i}\frac{\gamma}{\lambda}.\end{equation*}$$ Then, the energy saving due to pruning is $$\begin{align*} &\overline{\xi}= \prime{\xi} -\prime{\prime{\prime{\xi}}}= n*\rho_{t}\frac{\pi_{1}}{\beta}-\rho_{t}\frac{\pi_{2}}{\beta}+(n-1)*\rho_{i}\frac{\gamma}{\lambda}\end{align*}$$ $$\begin{align} \overline{\xi}= \frac{\rho_{t}}{\beta}[n*{\pi_{1}}-{\pi_{2}}]+(n-1)*\rho_{i}\frac{\gamma}{\lambda}. \end{align}$$(5) If energy is saved by pruning, then the necessary condition is |$\overline{\xi }> 0$|⁠. Thus, $$\begin{align*} &\frac{\rho_{t}}{\beta}[n*{\pi_{1}}-{\pi_{2}}]+(n-1)*\rho_{i}\frac{\gamma}{\lambda}> o \end{align*}$$ or $$\begin{align*} & \frac{n-1}{ [n*{\pi_{1}}-{\pi_{2}}]}> \frac{\frac{\rho_{t}}{\beta}}{\rho_{i}\frac{\gamma}{\lambda}}.\end{align*}$$ The right side of the above expression, |$\rho _{t}$|⁠, |$\beta $|⁠, |$\gamma $| and |$\lambda $| variables are positive values. $$\begin{align*} &\frac{(n-1)}{n*\pi_{1}-\pi_{2}}> 0. \end{align*}$$ Therefore, if |$(n*\pi _{1}-\pi _{2})\neq 0 $|⁠, then in pruning scenario, the mobile energy can be saved for the condition of |$n$| > 1. Also, if |$(n*\pi _{1}-\pi _{2}) < 0$|⁠, then energy must be saved. But, for the condition |$(n*\pi _{1}-\pi _{2})> 0$|⁠, the benefit of pruning will decrease in terms of energy. The condition of energy saving for pruning is that data transmission must be multiple or pruned data size in pruning condition > number of transmission |$\times $| transfer of data size in no pruning condition. The condition of energy saving for pruning is that data transmission must be multiple or pruned data size in pruning condition > number of transmission |$\times $| transfer of data size in no pruning condition. Open in new tab The condition of energy saving for pruning is that data transmission must be multiple or pruned data size in pruning condition > number of transmission |$\times $| transfer of data size in no pruning condition. The condition of energy saving for pruning is that data transmission must be multiple or pruned data size in pruning condition > number of transmission |$\times $| transfer of data size in no pruning condition. Open in new tab Time Saving: In no pruning condition, if |$n$| queries are sent to cloud from mobile, then it can be assumed that |$\pi _{1}$| amount of data is transferred for each query. According to Fig. 11, the total time in no pruning condition for the whole computation is |$t_{np}$|= |$t_{3}$|+|$t_{2}$|+|$t_{1}$| or $$\begin{align*} &t_{np}= n* [\frac{\gamma}{\mu}+ \frac{\pi_{1}}{\beta}+ \frac{\gamma}{\lambda}].\end{align*}$$ Here, |$n$| queries are processed first in mobile for |$t_{3}$| time and |$t_{2}$| time is required to send/receive |$\pi _{1}$| amount of data during each transmission and |$t_{3}$| time is needed for cloud processing. According to Figs 11 and 12, in pruning, the total time for whole computation is |$t_{p}$|= |$t_{3}$|+|$t_{2}$|+|$t_{1}$| or $$\begin{align*} &t_{p}= n*\frac{\gamma}{\mu}+ \frac{\pi_{2}}{\beta}+ \frac{\gamma}{\lambda}.\end{align*}$$ Hence, $$\begin{equation*}\frac{t_{np}}{t_{p}}= \frac{n* [\frac{\gamma}{\mu}+ \frac{\pi_{1}}{\beta}+ \frac{\gamma}{\lambda}]}{n*\frac{\gamma}{\mu}+ \frac{\pi_{2}}{\beta}+ \frac{\gamma}{\lambda}}\end{equation*}$$ or $$\begin{align} \frac{t_{np}}{t_{p}}= \frac{n}{1+\frac{(n-1)*\frac{\gamma}{\mu}+\frac{\pi_{2}}{\beta}-n*\frac{\pi_{1}}{\beta}}{\frac{\gamma}{\mu}+\frac{\pi_{1}}{\beta}+\frac{\gamma}{\lambda}} }.\end{align}$$(6) Always, |$\frac{t_{np}}{t_{p}}$| increases with the increase in |$n$|⁠. If |$((n-1)*\frac{\gamma }{\mu }+\frac{\pi _{2}}{\beta }-n*\frac{\pi _{1}}{\beta } < 0) $|⁠, then |$\frac{t_{np}}{t_{p}}$| increase. But, if |$((n-1)*\frac{\gamma }{\mu }+\frac{\pi _{2}}{\beta }-n*\frac{\pi _{1}}{\beta }> 0) $|⁠, then the value of |$\frac{t_{np}}{t_{p}}$| decreases. 4 Results and Discussion Mainly, three groups of analysis results are presented in this section. Analysis related to transmission of data (pruned data) with single query to cloud. Analysis of energy saving in three scenarios (no pruning, partial pruning and partial pruning). Analysis of the pruning gain in terms of energy and delay. 4.1 Analysis for pruning data An analytic model is constructed in Section 3.1 where the amount of energy consumption for pruning data in response to a single query to cloud is represented using Equation 1. From Equation 1, it is evident that the amount of pruned data is directly related to the bandwidth of the wireless network, transmission power, mobile processor speed and the energy of the mobile devices. These are affected by the number of transmitted messages. Also, the efficiency of the application depends on the processing speed of the cloud or server processor in comparison with the mobile device processor (then, we can consider as factor |$\psi $|⁠). Equation |$1$| is recalled here: $$\begin{align*} & \pi_{2}= \frac{\beta}{\rho_{t}}[\xi-\frac{\gamma}{\mu}(\rho_{m}+\frac{\rho_{i}}{\psi})]. \end{align*}$$ To realize the issues related to mobile-server communication, we have experimented with Nexus S |$GT-i9020A$| mobile with |$1$| GHz Single-Core ARM Cortex-A8 processor and |$8$| core servers with |$3.2$| GHZ. The values for |$\rho _{t}$| can be changed, but |$\rho _{i}$| and |$\rho _{m}$| are fixed for a particular mobile. The values of energy consumption of Nexus S mobile are |$\rho _{i}=0.9$| watt for idle position and processing energy |$\rho _{m}$|= 0.99=1.0 watt. Then, the speed up (⁠|$\psi $|⁠) of cloud server in this scenario is $$\begin{align*} &\psi=\frac{\lambda}{\mu}=\frac{3.2*8*1024}{1024}=26.\end{align*}$$ The normalized equation for the general equation is $$\begin{align*} & \pi= \frac{\beta}{\rho_{t}}[\xi-\frac{\gamma}{\mu}(1+\frac{0.9}{26})] [(1+\frac{0.9}{26})=1] \end{align*}$$ or $$\begin{align} \pi= \frac{\beta}{\rho_{t}}[\xi-\frac{\gamma}{\mu}]. \end{align}$$(7) (Here, |$\gamma /\mu $| is negligible to |$\gamma $|⁠.) $$\begin{align} \pi=\frac{\beta}{\rho_{t}}*\xi. \end{align}$$(8) During the mathematical analysis, the following issues are considered based on Equation |$1$|⁠: amount of data transfer with varying bandwidth and specific amount of mobile energy consumption, amount of mobile energy consumption for increase in bandwidth for fixed amount of pruned data, amount of data transfer from cloud server to mobile device for different transmission power settings with varying energy consumption of mobile devices, amount of data transfer between cloud server and mobile devices with varying transmission power while keeping the mobile energy consumption fixed and amount of energy consumption of mobile device for varying amount of data transfer from cloud to mobile device with varying bandwidth. In Fig. 13, it can clearly be seen that for increased bandwidth, amount of pruned data increases with varying mobile energy consumption. Figure 13. Open in new tabDownload slide Pruned data with bandwidth for the fixed of energy consumption. Figure 13. Open in new tabDownload slide Pruned data with bandwidth for the fixed of energy consumption. Figure 14 shows that for fixed amount of pruned data, if bandwidth increases then mobile energy consumption decreases. In case of low bandwidth, data transfer or pruning must be avoided due to high energy consumption. However, for fixed amount of pruned data, more transmission energy is required at low bandwidth that directly impacts the limited battery-powered resources. Figure 14. Open in new tabDownload slide Mobile energy consumption with bandwidth for the fixed of pruned data. Figure 14. Open in new tabDownload slide Mobile energy consumption with bandwidth for the fixed of pruned data. Figure 15 shows the relation between mobile energy consumption with amount of pruned data under different transmission power settings. It signifies that when energy consumption is not an issue, higher amount of data can be pruned. But for increase in transmission power settings the amount of pruned data will decrease. This is also evident from Fig. 16, which validates Equation 5. With fixed bandwidth and maintaining the same energy condition, if we increase transmission power, less amount of data can be pruned. Figure 15. Open in new tabDownload slide Pruned data with energy consumption for the fixed transmission power. Figure 15. Open in new tabDownload slide Pruned data with energy consumption for the fixed transmission power. Figure 16. Open in new tabDownload slide The relation between pruned data with transmission power for the fixed of mobile energy. Figure 16. Open in new tabDownload slide The relation between pruned data with transmission power for the fixed of mobile energy. Figure 17 signifies that if there is a requirement for pruning more data, the mobile energy consumption will increase. Thus, in this section, our analytic model is validated with results. From the above set of experiments, following conclusions may be drawn. Figure 17. Open in new tabDownload slide Pruned data with energy consumption for the fixed bandwidth. Figure 17. Open in new tabDownload slide Pruned data with energy consumption for the fixed bandwidth. Observations: if bandwidth is low in certain areas, data transfer or pruning must be avoided to reduce energy consumption; if there is any requirement for pruning large amount of data, bandwidth of the network should be increased and issue of energy-constrained mobile devices should be addressed; for pruning large amount of data from cloud to mobile device, mobile transmission power settings must be kept as low as possible. 4.2 Analysis for pruning gain With Equations 2–4, the energy consumption can be estimated for the three scenarios, no pruning, partial pruning and pruning. In Equation 5, if the number of queries sent to cloud is varied, energy consumption increases for the same amount of downloaded data as shown in Fig. 18. Figure 18. Open in new tabDownload slide Increasing energy with fixed data size for increasing number of queries. Figure 18. Open in new tabDownload slide Increasing energy with fixed data size for increasing number of queries. In Scenario |$2$|⁠, total number of |$n=1000$| queries are sent to the server and |$k=100$| queries are processed on the mobile device, i.e. partial pruning is performed. It is observed from Equations |$7$|–|$9$| that the energy consumption on mobile devices depends on the queries sent to the cloud or server, i.e. number of packet transmission between the mobile device and server or cloud. In Fig. 19, the comparison of the three cases with |$n=1000$| and |$k=100$| are shown. It is observed that the mobile device consumes less energy for Scenario 3. In Scenario 3, with the help of a pruning algorithm, data are pruned from the server by a single query and brought to the mobile device. Thereafter, all the queries are processed on the mobile device only. Therefore, number of packet transmission is minimized in Scenario 3. But it is noticed that if more number of queries are sent to the cloud, the number of packet transmission increases and the mobile device consumes more energy. As a result mobile battery power is wasted drastically. From the above results, conclusions can be drawn. Observations: pruning is the best choice when mobile energy consumption is considered; however, if storage on mobile device is a constraint, partial pruning can also be considered. Figure 19. Open in new tabDownload slide Energy consumption and pruned data for the three scenarios. Figure 19. Open in new tabDownload slide Energy consumption and pruned data for the three scenarios. Pruning gain analysis for energy saving. It has already been discussed in Section 3 that energy and time both are saved with the use of pruning. We also find the criteria for pruning with the help of Equations 5 and |$6$|⁠. Pruning gain can be an important measure in terms of mobile energy and time saving for taking decision regarding pruning. Pruning gain can be defined in terms of percentage of mobile energy savings due to pruning. Pruning gain can be derived from the Equation |$5$|⁠. The energy gain with pruning can be analyzed as follows: the variation in pruning gain in terms of energy saving for varying transmission power with different network bandwidths; the variation in pruning gain in terms of energy saving with varying data sizes; the variation in energy and network bandwidth with different transmission power settings. The earlier specified experimental set up (the Nexus S |$GT-i9020A$| mobile with |$1$| GHz Single-Core ARM Cortex-A8 processor and |$8$| core servers with |$3.2$| GHZ, |$\rho _{t}$|=1.5, |$n=1000$|⁠) has been used for the analysis purpose. Also, it is assumed that |$\pi _{1}= \pi _{2}$| and |$F=\frac{\pi _{1}}{\pi _{2}}$| for the analysis. The results of the analysis are shown in Figs 20– 22. Figure 20. Open in new tabDownload slide Energy gain with transmission power for the fixed bandwidth. Figure 20. Open in new tabDownload slide Energy gain with transmission power for the fixed bandwidth. Figure 21. Open in new tabDownload slide Pruning gain as energy saving with data size factor on fixed bandwidth. Figure 21. Open in new tabDownload slide Pruning gain as energy saving with data size factor on fixed bandwidth. Figure 22. Open in new tabDownload slide The relation between pruning gain as energy saving with bandwidth in fixed transmission power. Figure 22. Open in new tabDownload slide The relation between pruning gain as energy saving with bandwidth in fixed transmission power. In Fig. 20, it has been shown that for increasing transmission power of mobile device, the pruning gain (percentage of energy gain) increases in terms of energy. But, with an increase in bandwidth, the pruning gain decreases and, in certain cases, it can be negative; hence, no pruning becomes more advantageous than pruning in such cases. Hence, to get the maximum pruning gain, both transmission power and bandwidth of the transmission medium should be optimized. Figure 21 shows that with pruning, energy gain decreases for a fixed bandwidth with increase in the data size. Even in certain cases, it tends to zero implying that for a fixed bandwidth, no energy gain can happen with increase in data size. The pruning gain in terms of energy saving decreases with increase in bandwidth for a fixed transmission power setting as shown in Fig. 22. Pruning gain analysis for time saving. Earlier, pruning gain has been defined and it has been shown that time can be saved in this scenario in comparison with no pruning condition. In Equation 6, the pruning gain in terms of time is shown. The time gain with respect to pruning can be analyzed as follows. How much pruning gain in terms of time saving can be achieved for varying bandwidth with fixed number of packets transmission? How much pruning gain in terms of time saving can be achieved for varying number of packet transmission with fixed bandwidth? The above-mentioned analysis results are shown in Figs 23 and 24. Figure 23. Open in new tabDownload slide Pruning gain as time saving for the bandwidth in fixed number of transmission. Figure 23. Open in new tabDownload slide Pruning gain as time saving for the bandwidth in fixed number of transmission. Figure 24. Open in new tabDownload slide Pruning gain as time saving for the number of transmission in fixed bandwidth. Figure 24. Open in new tabDownload slide Pruning gain as time saving for the number of transmission in fixed bandwidth. With fixed number of packet transmissions, if bandwidth increases, then pruning gain in terms of time saving decreases because the time required to prune the data in the pruning case increases and becomes closer to no pruning case, and in such a situation, no benefit is observed due to pruning. This is shown in Fig. 23. But, with a fixed bandwidth, with and increase in number of packet transmission, time saving can be increased as shown in Fig. 24. Earlier, Equations 2–4 are validated with these results. Observation: pruning is a solution to the problem of high energy consumption, as well as time delay issues for communication between mobile devices and cloud; pruning gain in terms of energy consumption increases when bandwidth is low and mobile power setting is kept as high as possible; pruning gain in terms of time saving for multiple transmissions increases when bandwidth is low. 5 Conclusion In this paper, analytical models have been developed to decide whether a subgraph from a large, complex graph stored in cloud should be pruned and brought to mobile devices in order to reduce energy consumption of the mobile devices and also to reduce time delay in communication and processing. Three scenarios have been evaluated, such as no pruning, partial pruning and pruning for the implementation of a scheme for mobile assisted remote healthcare delivery. The decision-making process involves figuring out whether to perform repeated queries to get responses from the cloud data storage or to download the subgraph to the mobile devices for local processing. Mobile energy or battery power and transmission delay between mobile and cloud are chosen as criteria for the evaluation of the three scenarios. The analytical model is built based on different characteristics and parameters related to the mobile devices, cloud and network properties. It is shown that pruning is an optimal solution for our healthcare scheme in terms of delay and energy consumption of mobile devices. It is also shown that the amount of pruned data is directly dependent on the bandwidth of the transmission medium, transmission power and energy of the mobile devices. One important measure, pruning gain is evaluated in terms of mobile energy and time saving and how the parameter is affected due to the variation in bandwidth and number of message transmissions is also shown. Though the analytical model presented in this paper provides some indication regarding effective use of pruning algorithms in a remote healthcare scheme, the model can be effectively used for other mobile- and cloud-based applications as well. Further, in this paper, only analytical model has been presented. There is a need for real-life use cases or experiments to analyse the energy consumption and time delay in real-life scenarios in mobile and cloud collaboration and to explore the advantages of implementation of pruning algorithms while handling big data. We also believe that pruning a subgraph for a remote healthcare scheme will assist rapid and accurate diagnosis of diseases in mobile-assisted healthcare delivery scheme. Real-life experiments in order to demonstrate this will be carried out as our future work. Several results presented in this research tend to converge to optimization problems involving different parameters of mobile devices, as well as communication networks in pruning solutions, which may be addressed in future. DATA AVAILABILITY Data sharing is not applicable to this article as no new data were created or analyzed in this article. References [1] Mondal , S. and Mukherjee , N. ( 2016 ) Mobile-Assisted Remote Healthcare Delivery . In 2016 Fourth Int. Conf. Parallel, Distributed and Grid Computing (PDGC) , pp. 630 – 635 . IEEE , Waknaghat, India . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [2] Mondal , S. and Mukherjee , N. ( 2017 ) A Framework for ICT-Based Primary Healthcare Delivery for Children . In 2017 9th Int. Conf. Communication Systems and Networks (COMSNETS) , pp. 525 – 529 . IEEE , Bangalore, India . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [3] Mondal , S. and Mukherjee , N. ( 2018 ) A BFS Based Pruning Algorithm for Disease-Symptom Knowledge Graph Database . In 2018 3rd Int. Conf. Information and Communication Technology for Intelligent Systems (ICTIS 2018) , April 6–7 . IEEE , Bangalore, India . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [4] Das , R. , Mondal , S. and Mukherjee , N. ( 2018 ) More-Care: Mobile-Assisted Remote Healthcare Service Delivery . In 2018 10th Int. Conf. Communication Systems Networks (COMSNETS) , pp. 677 – 681 . IEEE [5] Mondal , S. ( 2019 ) An efficient reachability query based pruning algorithm in e-health scenario . J. Biomed. Inform. , 94 , 103171 . Google Scholar Crossref Search ADS PubMed WorldCat [6] Zaman , N. and Almusalli , F.A. ( 2017 ) Review: Smartphones Power Consumption Energy Saving Techniques . In 2017 Int. Conf. Innovations in Electrical Engineering and Computational Technologies (ICIEECT) , pp. 1 – 7 . IEEE , Karachi, Pakistan . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [7] Pramanik , P.K.D. ( 2019 ) Power consumption analysis, measurement, management, and issues: A state-of-the-art review of smartphone battery and energy usage . IEEE Access , 7 , 182113 – 182172 . Google Scholar Crossref Search ADS WorldCat [8] Carroll , A. and Heiser , G. ( 2010 ) An Analysis of Power Consumption in a Smartphone . In Proc. 2010 USENIX Conf. USENIX Annual Technical Conf., USENIXATC’10 , pp. 21 – 21 . USENIX Association , Berkeley, CA, USA . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC [9] Perrucci , G.P. , Fitzek , F.H.P., Sasso , G., Kellerer , W. and Widmer , J. ( 2009 ) On the Impact of 2G and 3G Network Usage for Mobile Phones’ Battery Life . In 2009 European Wireless Conf. , pp. 255 – 259 . IEEE , Aalborg, Denmark . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [10] Yan , M. ( 2019 ) Modeling the total energy consumption of mobile network services and applications . Energies , 12 , 184 . Google Scholar Crossref Search ADS WorldCat [11] Ahmad , R.W. A survey on energy estimation and power modeling schemes for smartphone applications . Int. J. Commun. Syst. , 30 , e3234 . Crossref Search ADS WorldCat [12] Tawalbeh , M. ( 2016 ) Studying the energy consumption in mobile devices . Procedia Computer Science , 94 , 183 – 189 . The 11th International Conference on Future Networks and Communications (FNC 2016)/The 13th International Conference on Mobile Systems and Pervasive Computing (MobiSPC 2016)/Affiliated Workshops . Google Scholar Crossref Search ADS WorldCat [13] Rice , A. and Hay , S. ( 2010 ) Decomposing Power Measurements for Mobile Devices . In 2010 IEEE Int. Conf. Pervasive Computing and Communications (PerCom) , pp. 70 – 78 . IEEE , Mannheim, Germany . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [14] Xiao , Y. , Savolainen , P., Karppanen , A., Siekkinen , M. and Ylä-Jääski , A. ( 2010 ) Practical Power Modeling of Data Transmission over 802.11g for Wireless Applications . In Proc. 1st Int. Conf. Energy-Efficient Computing and Networking, e-Energy ‘10 , pp. 75 – 84 . ACM , New York, NY, USA . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [15] Zhang , L. , Tiwana , B., Dick , R.P., Qian , Z., Mao , Z.M., Wang , Z. and Yang , L. ( 2010 ) Accurate Online Power Estimation and Automatic Battery Behavior Based Power Model Generation for Smartphones . In 2010 IEEE/ACM/IFIP Int. Conf. Hardware/Software Codesign and System Synthesis (CODES+ISSS) , pp. 105 – 114 . IEEE , Scottsdale, AZ, USA . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [16] Xiao , Y. , Bhaumik , R., Yang , Z., Siekkinen , M., Savolainen , P. and Yla-Jaaski , A. ( 2010 ) A System-Level Model for Runtime Power Estimation on Mobile Devices . In 2010 IEEE/ACM Int. Conf. Green Computing and Communications Int. Conf. Cyber, Physical and Social Computing , pp. 27 – 34 . IEEE , Scottsdale, AZ, USA . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [17] Gu , X. ( 2004 ) Adaptive offloading for pervasive computing . IEEE Pervasive Comput. , 3 , 66 – 73 . Google Scholar OpenURL Placeholder Text WorldCat [18] Chen , G. ( 2004 ) Studying energy trade-offs in offloading computation/compilation in java-enabled mobile devices . IEEE Trans. Parallel Distrib. Syst. , 15 , 795 – 809 . Google Scholar Crossref Search ADS WorldCat [19] Kumar , K. ( 2010 ) Cloud computing for mobile users: Can offloading computation save energy? Computer , 43 , 51 – 56 . Google Scholar Crossref Search ADS WorldCat [20] Han , S. , Zhang , S. and Zhang , Y. ( 2006 ) Energy Saving of Mobile Devices Based on Component Migration and Replication in Pervasive Computing . In Proc. Third Int. Conf. Ubiquitous Intelligence and Computing, UIC ‘06 , pp. 637 – 647 . Springer , Berlin, Heidelberg . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [21] Hong , Y. , Kumar , K. and Lu , Y. ( 2009 ) Energy Efficient Content-Based Image Retrieval for Mobile Systems . In 2009 IEEE Int. Symposium on Circuits and Systems , pp. 1673 – 1676 . IEEE , Taipei, Taiwan . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [22] Li , Z. , Wang , C. and Xu , R. ( 2001 ) Computation Offloading to Save Energy on Handheld Devices: A Partition Scheme . In Proc. 2001 Int. Conf. Compilers, Architecture, and Synthesis for Embedded Systems, CASES ‘01 , pp. 238 – 246 . ACM , New York, NY, USA . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [23] Cuervo , E. , Balasubramanian , A., Cho , D.-k., Wolman , A., Saroiu , S., Chandra , R. and Bahl , P. ( 2010 ) MAUI: Making Smartphones Last Longer with Code Offload . In Proc. 8th Int. Conf. Mobile Systems, Applications, and Services, MobiSys ‘10 , pp. 49 – 62 . ACM , New York, NY, USA . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [24] Seshasayee , B. , Nathuji , R. and Schwan , K. ( 2007 ) Energy-Aware Mobile Service Overlays: Cooperative Dynamic Power Management in Distributed Mobile Systems . In Fourth Int. Conf. Autonomic Computing (ICAC ‘07) , pp. 6 – 6 . IEEE , Jacksonville, FL, USA . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [25] Malik , S.U.R. , Akram , H., Gill , S.S., Pervaiz , H. and Malik , H. ( 2020 ) EFFORT: Energy efficient framework for offload communication in mobile cloud computing . Softw. Pract. Exp., 1–14. https://doi.org/10.1002/spe.2850. Google Scholar OpenURL Placeholder Text WorldCat [26] Wolski , R. , Gurun , S., Krintz , C. and Nurmi , D. ( 2008 ) Using Bandwidth Data to Make Computation Offloading Decisions . In 2008 IEEE Int. Symposium on Parallel and Distributed Processing , pp. 1 – 8 . IEEE , Scottsdale, AZ, USA . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [27] Miettinen , A.P. and Nurminen , J.K. ( 2010 ) Energy Efficiency of Mobile Clients in Cloud Computing . In Proc. 2nd USENIX Conf. Hot Topics in Cloud Computing, HotCloud ‘10 , p. 4 . USENIX Association , Berkeley, CA, USA . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC [28] Wang , C. ( 2004 ) A computation offloading scheme on handheld devices . J. Parallel Distrib. Comput. , 64 , 740 – 746 . Google Scholar Crossref Search ADS WorldCat [29] Sardellitti , S. ( 2015 ) Joint optimization of radio and computational resources for multicell mobile-edge computing . IEEE Trans. Signal Inf. Process. Netw. , 1 , 89 – 103 . Google Scholar OpenURL Placeholder Text WorldCat [30] Yao , F. , Demers , A. and Shenker , S. ( 1995 ) A Scheduling Model for Reduced CPU Energy . In Proc. IEEE 36th Annual Foundations of Computer Science , pp. 374 – 382 . IEEE , Milwaukee, WI, USA . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC [31] Yuan , W. ( 2003 ) Energy-efficient soft real-time CPU scheduling for mobile multimedia systems . SIGOPS Oper. Syst. Rev. , 37 , 149 – 163 . Google Scholar Crossref Search ADS WorldCat [32] Benini , L. ( 1999 ) Policy optimization for dynamic power management . IEEE Trans. Comput. Aid Design Integr. Circuits Syst. , 18 , 813 – 833 . Google Scholar Crossref Search ADS WorldCat [33] Zhang , W. ( 2013 ) Energy-optimal mobile cloud computing under stochastic wireless channel . IEEE Trans. Wirel. Commun. , 12 , 4569 – 4581 . Google Scholar Crossref Search ADS WorldCat [34] Pihkola , H. ( 2018 ) Evaluating the energy consumption of mobile data transfer-from technology development to consumer behaviour and life cycle thinking . Sustainability , 10 , 2494. Google Scholar OpenURL Placeholder Text WorldCat © The British Computer Society 2021. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - Pruning of Health Data in Mobile-Assisted Remote Healthcare Service Delivery
JF - The Computer Journal
DO - 10.1093/comjnl/bxab083
DA - 2021-06-24
UR - https://www.deepdyve.com/lp/oxford-university-press/pruning-of-health-data-in-mobile-assisted-remote-healthcare-service-ExJN3d3hY2
SP - 1
EP - 1
VL - Advance Article
IS - 
DP - DeepDyve
ER -