TOPSIS-Based Service Arbitration for Autonomic ... - IEEE Xplore

Received February 11, 2016, accepted February 27, 2016, date of publication March 23, 2016, date of current version April 12, 2016. Digital Object Identifier 10.1109/ACCESS.2016.2545741

TOPSIS-Based Service Arbitration for Autonomic Internet of Things QAZI MAMOON ASHRAF1 , (Member, IEEE), MOHAMED HADI HABAEBI2 , (Member, IEEE), and MD. RAFIQUL ISLAM2 , (Member, IEEE) 1 Telekom

Malaysia Research and Development, Cyberjaya 63000, Malaysia of Electrical and Computer Engineering, University Islam Antarabangsa, Kuala Lumpur 53100, Malaysia

2 Department

Corresponding author: Q. M. Ashraf ([email protected]) This paper has supplementary downloadable material available at http://ieeexplore.ieee.org., provided by the author.

ABSTRACT Recent research on Internet of Things (IoT) has focused on the adaptation of the autonomic computing paradigm to make IoT self-sufficient. Service arbitration is one aspect which can greatly benefit from the adoption of the autonomic theory. Instead of allowing all deployed devices to be active, only a selected set of devices can be utilized to provide a particular service. This paper proposes a dynamic service arbitration scheme for this purpose. The approach for the service arbitration scheme is based on the technique for order preference by similarity to ideal solution (TOPSIS) algorithm. This method supplements existing autonomic frameworks with the aim to minimize user intervention as well as imparting self-configuration in the system. The analysis through TOPSIS can be extended to any number of permutations and combinations of alternatives and system policies. INDEX TERMS Internet of things, scalability, wireless sensor networks, autonomy, self-configuration, TOPSIS. I. INTRODUCTION

The management of devices and data is an important research problem in the Internet of Things (IoT). One approach to solving the management problem is to apply the theory of autonomic computing to aid decision making in the management process [1]. The application of the autonomic control loop should, in theory, make IoT self-sufficient. Recently, [2], [3] have attempted to employ concepts of autonomic computing in IoT. However, as [3] point out, full autonomy is not currently possible in IoT. Currently, what exists is partial autonomy, which is based on the fulfillment of a particular set of autonomic requirements. Devices in IoT, ranging from smart phones to embedded sensor devices, fulfill three important requirements of computation, organization and communication. These three requirements are essential to allow for device connectivity in IoT [4]. A question that arises is concerned with the distribution of ownership of the data, the devices as well as the services for a complex IoT ecosystem. This complexity, arising because of the different owners of services, becomes a big stumbling block in implementation efforts [5]. Such complexity will defeat the purpose of IoT to fuse different data sources. As an analogy, this is similar to the practice of building cellular base stations at the same location which

VOLUME 4, 2016

provides same coverage to users. Investment could be highly minimized if the same cellular towers are re-used across various cellular operators. Such a solution is highly cost efficient due to the presence of hundreds of such sensors in IoT as replication of same sensors at the same locations by each IoT company is impractical. The problem that arises now is to solve the manner in which same sensors can provide services to different stakeholders. To mitigate this confusion, the concept of sensing as a service (SaaS) was discussed by Zaslavsky et al. in 2013 [6]. SaaS is implemented through an owner of sensing elements who provides data services on demand for a small fee or discount. The owners could be either ‘‘a person, private organization, public organization, or government’’ [6]. Not having a single owner complicates the IoT ecosystem tremendously, and it is important to deal with just the data instead of the owners. In implementation of smart street lights, for instance, the system supports a large number of different devices mounted on street lights and connected on a mesh network. In a typical setting, the city owns the street assets and pays for the electricity, a third party installs and maintains the lights, and the management is sourced to somewhere else. Combining different data and systems to create new services is going to be an important creator of innovative

2169-3536 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

1313

Q. M. Ashraf et al.: TOPSIS-Based Service Arbitration for Autonomic Internet of Things

businesses and new revenues for the infrastructure providers. SaaS solves this by exposing the data as application program interfaces (APIs) [7]. Perhaps a third party utility may want to pay the city to put a new sensor up on the street lighting infrastructure. The communication network may provide services to others that the city has no involvement in - perhaps smart meters. Ideally, each bit of the infrastructure should be treated as a service. That takes a bit of a shift in approach by the city and the companies servicing the infrastructure. To achieve management of such a large number of devices, offering different services becomes a difficult task due to the problems highlighted above. In addition, there is the issue of scalability. Scalability is defined differently in different contexts. For the case of IoT, scalability is the ability of a system to not undergo significant loss in performance when subjected to an increased load, and increased number of constituent elements [8]. Furthermore, it is extremely inefficient to allow all constituent end devices1 to be active all the time to provide for a general set of services. Dynamic management of sleeping cycles in the devices allows for a more efficient solution. In this method, some devices are sent to a low power state temporarily while other devices are kept in the active state [9]. In real time systems, such a decision cannot be made through human contact and this decision will have to be assigned to a computer. This is where autonomic computing can play its part in decision making. Various IoT devices are capable of providing different sensing and routing functionalities. Autonomic computing will simplify the complexity involved to arrive at a decision with all the variables in the scene. The autonomic computing paradigm provides a general management scheme that caters for the requirements from all the devices [10]. Also, currently, there is no standardized autonomic scheme that allows the system to harness the value of heterogeneity for management [11]. In the example above, the duty cycle decision will have to be based on the resources provided by the IoT devices. For example, some devices may have higher battery energy levels whereas other devices may be running out of battery energy. Such variety demands that both sets of devices be treated differently. Thus, a decision will have to be made to assign different duty cycles based on the current context to maximize operational or sensing efficiency. It is also essential to keep track of their status as well as details at a central management entity which further complicates device management [12]. The most important contribution of this paper is a scheme for service arbitration and selection for use in IoT. This contribution is a solution for 1) the problem of service arbitration in IoT, 2) using Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) numerical modeling method, as well as 3) implementing the concepts of autonomic computing. The methodology is demonstrated for duty cycle decision making for heterogeneous set of devices. This paper is 1 The terms device, mobile node and end node may be used interchangeably. Devices may be heterogeneous in nature in terms of supported sensor measurements as well as functional capabilities. An ideal system should cater for such heterogeneity.

1314

organized as follows. Section II presents the preliminaries and Section III presents the related work and the motivation behind this research. Section IV presents the proposed scheme and its analysis. Section V presents the conclusion. II. PRELIMINARIES A. AUTONOMIC CONTROL LOOP

The autonomic control loop provides the foundation of the theory of autonomic computing [13]. It provides the definition of the essential functions that are needed in an autonomic system. Thus, implementing the theory of autonomic computing in the service arbitration scheme means to follow the style and architectural elements suggested by the autonomic control loop: the elements to monitor, analyze, plan and execute (MAPE). These four elements have been discussed next on the basis of their functionality in the context of the service arbitration process: 1) THE MONITOR MODULE

As the name suggests, this module is responsible for generating the input to any autonomic system, and essentially observes parameters of the environment. This allows the autonomic system to be aware about its state, and to observe any possible changes. Service arbitration can greatly benefit from a wealth of parameters that can be monitored in terms of device states as well as environment states [14]. Monitoring properties such as energy remaining, memory resources, as well as sensor availability in the resources in a network can allow innovative decision making. The more variables that are monitored, the greater is the impact on decision making. 2) THE ANALYZE MODULE

The monitoring phase results in the collection of a large amount of control and sensor data. The analyze module works on this collected data to derive benefit for the autonomic system. Essentially, for the proposed service arbitration method, the analysis is performed through the application of the TOPSIS scheme which will be discussed later. The main responsibility is to provide a mechanism to select the most appropriate alternative for a set of devices. A proper analysis suitable under the applied context will allow the system to manage its resources in the most efficient way. 3) THE PLAN MODULE

The responsibility of the plan module is to highlight the goal of the system or scheme. For example, in our context, the goal is to maximize the network efficiency by implementation of a dynamic duty cycle scheme. The plan module sets down guidelines and targets for the scheme to achieve. Over time, the goals may keep on changing, and the autonomic system has to acknowledge the changes. In our scheme, this is achieved by varying the parameter weights in the TOPSIS algorithm. The plan module works with higher level, user defined policies, rules and regulations which are basically system level constraints. VOLUME 4, 2016


4) THE EXECUTE MODULE

The execute module controls the execution of the formulated plan. It is also responsible for offering feedback to the monitor module. In our scenario, execution of the plan takes effect in the form of self-configuration of the duty cycle.

Algorithm 1 Example for Static Service Selection at Autonomic Manager 1. [X,Y,Z] ← 2. transmitPower ← X 3. dim(Y ) 4. dutyCycle ← Z

B. AUTONOMIC SYSTEM MODEL FOR IoT

This work is motivated by the self-configuration scheme previously proposed in [2]. The scheme attempts to achieve management of resource constrained devices using service arbitration. The design is modular in nature as the functionality is divided between a hierarchy of layers based on their respective roles, functionalities and capabilities. Similar to the use case previously discussed in [15], we assume that a gateway device takes the role of the autonomic manager (AM) and the IoT devices fulfill the role of the managed devices (MD). This has been exhibited in Fig. 1. AMs are the basic building blocks of autonomic systems, and the gateway is responsible for the analysis through TOPSIS. In a high level overview, they can be considered as services within service oriented architectures, and can simultaneously be providing and consuming services [16].

web interface. Here, the service levels required are set manually according to plan. 2. In dynamic service arbitration, the higher level policy, as well as the current service levels, are checked first, and then the AM selects the service parameters appropriately. In our proposed design, the ‘‘Monitor’’ module will allow gathering of current service levels as well as the sensor data for feedback. The ‘‘Plan’’ module will check the requirements in the higher level policy, such as the number of temperature readings needed from the network in one minute, or the percentage of electricity to be saved. The ‘‘Analyze’’ module will compute the alternatives using the TOPSIS model. Finally, the ‘‘Execute’’ module will deliver service information accordingly and configure the duty cycle of the IoT devices. Algorithm 2 summarizes the dynamic service arbitration example. Algorithm 2 Example for Dynamic Service Selection Set by a Function in Autonomic Manager (AM) 1. [X,Y,Z] ← f(AM) 2. transmitPower ← X 3. dim(Y ) 4. dutyCycle ← Z III. RELATED WORK A. SERVICE SELECTION SCHEMES

FIGURE 1. Hierarchy of layers in system under consideration. A gateway device takes the role of the autonomic manager (AM) whereas IoT devices fulfill the role of the managed device.

C. SERVICE ARBITRATION

Before going into the description of the scheme, it is important to understand the concept related to service arbitration that is being applied here. This definition may vary with other definitions for service based computing such as those provided for web services and related architectures. We define service arbitration as ‘‘the selection of system resources such that the system as a whole meets a specific set of requirements to provide and consume specific services’’. Subsequently, service arbitration can be either static or dynamic: 1. In static service arbitration, the user selects a value for a set of parameters such as the recommended dimming level for an IoT street light, duty cycle for a temperature sensor or transmit power for an IoT device as shown in Algorithm 1. These values are obtained from the user via appropriate interfaces from the application such as a middleware or a VOLUME 4, 2016

Service arbitration has been used in a variety of wireless devices to impart network dynamism as well as improved management. It is considered to be the future of intelligent computing in IoT. Many different proposals of service arbitration mechanisms based on context and state parameters have emerged and new ones are still being published. Recently, [17] presented CASSARAM to address the research challenges of selecting sensors when large numbers of sensors with overlapping and redundant functionality are available. CASSARAM proposes the search and selection of sensors based on user priorities. CASSARAM considers a broad range of characteristics of sensors for search such as reliability, accuracy, and battery life, just to name a few. Sensors are ranked and indexed using a preset user based priority. CASSARAM is evaluated with the metrics of processing time and accuracy in percentage. However, the parameters to compare performance over a set of multiple nodes are not considered in this study. Furthermore, user interference is needed to specify priorities at higher layers. The same authors [18] present CA4IOT as a middleware IoT platform. CA4IOT automates the task of selecting the 1315


sensors according to the problems/tasks at hand. This is done to enable the paradigm of SaaS. The focus is on automated configuration using filtering and reasoning mechanisms. The decisions are made using higher level user problems. This method is similar to that being applied in CASSARAM [17]. However, the information regarding the best service is filtered out only at the higher level, and all of the nodes will continue to function at the highest service levels. The actual configuration of nodes does not take place which means that all nodes continue normal functioning. Therefore, the system is more concerned about data management than actual device management. Some works focus entirely on the scheduling and duty cycle service management. Research in [19] consider an adaptive energy efficient sensor scheduling mechanism to choose sets of active sensors to work alternatively to meet different type of queries. They consider simulation for up to 1000 sensors and work in multi-hop scenario. The queries are formed from a set of requirements. However, the difference between this study and ours is that their scheme is valid only for multi-hop scenarios whereas we are considering a single hop setup. Authors of [20] study the node sleep scheduling problem in the context of clustered sensor networks. They propose Linear Distance-based Scheduling (LDS) technique for sleeping in each cluster and evaluate using analytical procedures. The LDS scheme selects a sensor node to sleep with higher probability when it is situated far away from the cluster head. Authors of [21] in a filed patent state that the network can be efficiently configured if the configuration device reaches out to non-reporting nodes. The configuration device is required to do so when it does not expect to use its transmission resources to configure a reporting node. This can be accomplished by maintaining a list of nodes yet to be configured, using the node configuration and reporting times to determine periods in which the configuration device’s transmission resources will be idle, contacting non-reporting nodes during those periods. Authors of [22] propose a novel Fault Tolerant Service Selection Framework (FTSSF). They model various factors such as the jobs that can be completed by various service providers, and evaluate it using simulation methodology. They limit selection criteria to only 10, and consider up to 1000 nodes. Authors of [23] propose a Quality of Service (QoS) aware service selection approach. The approach first employs cloud model to compute the QoS uncertainty for pruning redundant services while extracting reliable services. Then, mixed integer programming is used to select optimal services. They compare the computation time to evaluate the performance. However, only the computation time is used to evaluate the performance of the service arbitration. There are no available measurements for other metrics such as change in duty cycle, or other QoS metrics. The authors of [24] propose a USPIOT (Ubiquitous Services Platform for IOT) for screening a variety of 1316

heterogeneous devices so that a variety of ubiquitous services can be achieved. They describe services using XML and use ANN (Artificial Neural Network) selection algorithm to find the best service. Evaluation metrics consider reliability and the number of learning. However, it is hard to compare the performance as the number of learning is specific for Neural Network based methods. It is not possible to evaluate our nonneural network based decision process in similar performance metrics. Authors of [25] study the resource virtualization and service encapsulation of a logistics center. They consider service selection/arbitration as an optimization problem and establish a Particle Swarm Optimization (PSO)-based web service selection model with QoS constraints. The feasibility and effectiveness of the model are verified by several experiments for fitness function and cost metrics. However, their concept of service selection deals with web services whereas our scope is limited to end resources contributing to overall service delivery. Some sort of abstraction can be enabled, but the service arbitration is residing in different networking layers. B. TOPSIS APPROACH TO DECISION MAKING

Multi attribute decision making methods (MADM) are extensively used to solve optimization problems. Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) is a MADM method and appears to be encouraging to be applied to the problem of service arbitration. TOPSIS was first developed by Yoon and Hwang [26] and Tzeng [27]. Typically, MADM techniques are based on a deterministic approach, used in decision making where results can be affected by several factors [28]. With the help of MADM, ranking of alternatives with respect to given criteria can be determined, and thus eventually can influence a decision. Different alternatives are defined by different attributes, and decisions are influenced by the weights associated with each attribute. The attributes which play a higher role in the decision making process are to be assigned a higher weight. In [29], the attributes used in the MADM process include signal-to-noise ratio, residency time, power level, security, moving speed and effective capacity to decide on the selection of neighbors for streaming. The key criteria is the path that the data will take, essentially the number of hops from source to the destination. Once a mobile node joins the network, its necessary details are collected by the gateway, similar to the registration scheme in [30]. It is important to realize that a single neighbor is not being selected. Instead, a set of the best neighbors is being selected. We apply a similar approach to select a set of best devices instead of selecting just a single device. We take the devices and divide them into many sets. Each set will have a different average/aggregate value for its properties. Then the best alternative set can be determined. Open Issues: Devices in IoT are highly dynamic and the same device may require providing different services in different network environment. For example, a dynamic node that is equipped with temperature and humidity sensors may only require providing temperature sensing functionality in VOLUME 4, 2016


one instance but both the sensing capabilities in another instance. The node may also be required to calculate average temperature in a day in one network environment but only transfer the temperature data in another network environment. These types of service flexibilities on a mobile node are highly anticipated in IoT. The open issues with these dynamic nodes to provide different services in different network environment can be summarized as follows: • It is difficult to anticipate the type of service that is required to be loaded at any given point of time. • Existing mobile nodes are not dynamic and automatic to be able to manipulate services without user intervention. IV. SERVICE ARBITRATION USING TOPSIS

In our context, service2 arbitration refers to the decision of choosing the most appropriate set of devices based on the parameters associated with the service and the available resources. As mentioned earlier, this decision takes place at the AM represented by the gateway. For our use case, the goal is to select the duty cycle for the devices connected to the gateway. Figure 2 shows the relationship between services and configuration parameters. One service may be defined by one or a multitude of configuration parameters.

Number of Neighbors. This parameter specifies measurements for the rounded average number of neighboring nodes within one hop of the constituent. • Memory Resources. This parameter highlights whether the set of devices have enough memory resources for loading new configuration parameters (such as by using over the air programming). This is a Boolean variable and can take only true or false values. For demonstration, we assume that the constraint policy in the plan module is that only 20 nodes with the highest energy levels and near to many nodes should be active at a given time. Service selection would refer to the dynamic decision of selecting the best set of devices according to these variables which is processed in the analysis module. Once the decision for the best alternative is made, the constituent devices will be configured to be active all of the time, whereas the other devices will be put to sleep. The service selection algorithm can allow dynamic decision making using feedback from the available resources in even real-time. Table 1 shows sample values that could be collected by the monitor module for six randomly generated alternatives containing 20 devices each. This data will be used to demonstrate the method and results later. •

TABLE 1. Attribute values for the alternative services.

A. DESIGN ASSUMPTIONS FIGURE 2. One service may need to configure one or multiple parameters under a one-to-many relationship.

Following the autonomic framework, the monitor module is responsible to collect attributes and state information. For the service selection problem, the following is a representative set of attributes that are considered in the decision making process using TOPSIS: • Energy Used. This is the measurement for the average amount of energy already used by the set of devices (expressed in percentage). • Sensing Frequency. This specifies how often the set of devices update the sensor readings. It is measured as total number of times per hour. 2 Services may refer to a broad array of functionalities arising in the end device such as frequency of sensing, frequency of update, duty cycle, sensor data type and security functionality. In middleware based perspectives, services refer to web application level functionalities that are provided by the Internet cloud. Services presented by devices will vary according to the host network and the available resources. VOLUME 4, 2016

1. There is not a specific general operating system present in the nodes. The framework and the service arbitration method can also be used by specific implementations which are looking to include autonomic self-configuration and benefit from the same. 2. The framework proposed in [2] basically provides for 1) device registration, 2) configuration of parameters upon device registration, 3) updating of configuration, as well as 4) configuration delivery. The service arbitration which was kept as future work is being discussed in this paper. 3. The alternatives will have different parameters which are represented by the set of Pn , where n differentiates the parameter. The plan module will assign each parameter a weight denoted by wn . Thus, an alternative can have n number of configuration parameters associated with it. In our example, P1 = EU, P2 = SF, P3 = NN, and P4 = MR. 4. For non-numerical parameters, the values can be translated to numerical values based on any suitable conversion scheme. As an example, Yes/No or True/False states can be represented in Boolean values. 1317


B. THE APPROACH FOR ANALYSIS

For simplicity in explaining, we are assuming six sets of alternatives, A1 - A6 containing 20 randomly selected devices each from a total of 50 devices. At the end of the decision, one of the alternatives will be selected and the remaining devices which are not in that alternative will be sent to sleep. Step 1: The first step is to construct the decision matrix with all alternatives and their monitored values: D = xkn  41  60   76 D=  42   29 84

56 60 89 64 53 57

21 22 13 21 22 10

 1 1  0  1  1 0

This step adds the squared values for a specific parameter Pn and finds the square root of the total. It then normalizes the decision variables by dividing them with this sum.   0.28484588 0.355521519 0.465518653 0.5 0.416847629 0.380915913 0.487686208 0.5   0.528006997 0.565025271 0.288178214 0   rkn =   0.291793341 0.406310307 0.465518653 0.5 0.201476354 0.336475723 0.443351099 0.5 0.583586681 0.361870117 0.221675549 0 Step 3: The next step is to decide on the relative importance of each of the attributes involved in the decision about service arbitration. For this purpose, each of the attribute is assigned a weight and the weighted decision matrix is constructed: 0.035552152 0.038091591 0.056502527 0.040631031 0.033647572 0.036187012

0.186207461 0.195074483 0.115271286 0.186207461 0.177340439 0.08867022

 0.05 0.05  0   0.05  0.05 0

Where, wn is the weight of criterion for the service level. We will select the two most important parameters i.e. energy used (EU) and number of neighbors (NN) according to the policy specified in the plan module. Each is assigned a weight of 0.4. Others are assigned a weight of 0.1, and thus, will also contribute to the decision making process. Thus, w1 = 0.4, w2 = 0.1, w3 = 0.4 and w4 = 0.1. Step 4: The next step is to find the best and the worst value for each of the attribute. For each of the alternative under consideration, the ideal solutions are determined. The positive ideal solution is a collection of the best values with 1318

Bkn 

 0.000438918 0.0000786241 0 0.000338963 0 0   0 0.00636855 0.0025  0.000251904 0.0000786241 0   0.000522349 0.000314496 0  0.00041272 0.011321867 0.0025   0.040368537  0.088093861     0.161022086    =   0.040443464   0.028928278  0.193896663

0.001112076 0.007421566  0.017059562 = 0.001305145   0 0.023361328

SPIS

Where, xkn is the value in the alternative k for the parameter n. Step 2: The next step is to normalize the decision matrix using the following formula: xkn rkn = qP 2 xkn

ϑkn = wn rkn  0.113938352 0.166739052  0.211202799 ϑkn =  0.116717336  0.080590542 0.233434672

respect to each criterion on the weighted normalized decision matrix while the negative ideal solution is a collection of the worst values in respect to each criterion. The separation of each alternative from the positive ideal solution is given below:

Similarly, the separation of each alternative from the negative ideal solution is give below: Wkn 

0.01427937  0.00444830   0.00049425 =  0.01362293   0.02336132 0.00000000

SNIS

0.0000036274 0.034673219 0.0000197493 0.038054054 0.0005223490 0.013287469 0.0000487687 0.034673219 0.0000000000 0.031449631 0.0000064487 0.007862408   0.226839628  0.21218414     0.119599642   =  0.22548819     0.239397076  0.088706576

 0.0025 0.0025   0   0.0025   0.0025  0

For each of the service alternatives under consideration (represented by a row in the matrix), its level of preference is measured. The preference level is measured in terms of distances from the best and the worst solutions. We divide the separation from the negative ideal solution by the summation of the positive and negative ideal solutions. The alternatives are ranked as follows: SNIS Rank = SPIS + S NIS   0.848924763  0.706625658     0.426195231   Rank =   0.847917827     0.892189546  0.313890868 Figure 3 summarizes the ranks for each of the alternatives. Clearly, alternative 5 comes out as a winner as being the closest to the positive ideal solution. The least favorable alternative turns out to be alternative 6, having consumed the VOLUME 4, 2016


TABLE 2. Results from weight distribution method.

FIGURE 3. Results for service alternative selection using weights of 0.4, 0.1, 0.4 and 0.1. Alternative 5 is the closest solution to the positive ideal solution, whereas alternative 6 is the closest to the negative ideal solution.

FIGURE 4. Results for service alternative selection using an equal weight of 0.25 for each. Alternative 4 is the closest solution to the positive ideal solution, whereas alternative 6 remains the closest to the negative ideal solution.

highest energy. Figure 4 displays results after re-performing the TOPSIS algorithm using equal weights of 0.25 for each parameters. In this case, alternative 6 still presents the worse choice. However, the candidate for the best choice is changed to alternative 4. The results indicate that even in case of different weights, the same alternative can be the worst choice, which needs to be avoided at all costs. Recalling the assumption earlier, the weights are determined solely by the system designer. This level of user intervention occurs only when the system designer wants to change the behavior of the system, and is therefore minimal. The user plays no part in the actual decision making analysis, and no human feedback is required. The application of TOPSIS model effectively finds the optimal balance between service demands, user demands, as well as available resources. The detailed calculations can be found in the supplementary file included with this manuscript. The results are benchmarked with a decision making method using the popular weighted distribution method. The same weights with the same values of variables are used. Table 2 highlights the results. For weight set 1, it can be seen that the ideal candidate is Alternative 3 with a score of 44.5, whereas the least ideal candidate is Alternative 5. Similarly for weight set 2, the ideal candidate is also Alternative 3, and the least ideal candidate is Alternative 5. Thus, the proposed scheme performs better in service arbitration related performs as it is more sensitive to change in weights. Using a powerful computational platform or hardware, the number of randomly generated alternatives can be extended to cover all permutations and combinations, and this algorithm can be run to obtain the best alternatives for service execution in IoT. This will help achieve management of a large number of devices where offering different services becomes a difficult task due to the problems highlighted earlier. Using such a VOLUME 4, 2016

dynamic selection scheme in real-time decreases the decision making time as the decision making process is moved from the user to the machine. As a result, more data sets with increasing number of permutations and combinations can be processed and analyzed. In our presented use case, it was assumed that it is extremely inefficient to allow all constituent end devices to be active all the time, and consuming energy. Some of these devices may have significant energy resources, whereas some of them may be at the verge of switching off. The proposed dynamic scheme was able to cater for such a situation, while keeping the higher level policies in mind as well. The state variables for all the elements were collected and organized into sets of alternatives. The choice of the number of alternatives is dependent on the system designer as well as the requirements of the context scenario. Similarly, MADM schemes, other than TOPSIS, can be used to come up with different analysis for dynamic service selection in the IoT. By supplementing the analysis with the framework of autonomic computing theory, IoT devices can actively take contribute to make the decision making process more efficient. The constant monitoring and feedback from the systems allows catering for the requirements from all the devices. It should be noted that it is not possible to objectively measure the accuracy of the rankings obtained by an MADM method. It is quite possible that different MADM algorithms will provide different rankings for alternatives with the same attribute values and decision criteria (weights). The TOPSIS method employed in this thesis, however, is capable of processing a large number of alternatives, and is of providing the best alternative in many scenarios. Upon its application, decision making for persistent actions is automated, and user intervention is kept at a minimum for the selection of the decision criteria. V. CONCLUSION

This paper proposed a dynamic service arbitration scheme for dynamic selection of services in an IoT setup. The approach for the service arbitration scheme is based on TOPSIS algorithm. This method supplements existing autonomic framework with the aim to minimize user intervention as well as imparting self-configuration. The analysis through TOPSIS can be extended to any permutations and combinations of alternatives and system policies. ACKNOWLEDGMENT

Q. M. Ashraf was with the Department of Electrical and Computer Engineering, University Islam Antarabangsa Malaysia. 1319


REFERENCES [1] L. Atzori, A. Iera, and G. Morabito, ‘‘The Internet of Things: A survey,’’ Comput. Netw., vol. 54, no. 15, pp. 2787–2805, Oct. 2010. [2] Q. Ashraf, M. Habaebi, G. R. Sinniah, M. Ahmed, S. Khan, and S. Hameed, ‘‘Autonomic protocol and architecture for devices in Internet of Things,’’ in Proc. IEEE Innov. Smart Grid Technol.-Asia (ISGT Asia), Kuala Lumpur, Malaysia, May 2014, pp. 737–742. [3] Q. M. Ashraf and M. H. Habaebi, ‘‘Autonomic schemes for threat mitigation in Internet of Things,’’ J. Netw. Comput. Appl., vol. 49, pp. 112–127, Mar. 2015. [4] A. Iera, C. Floerkemeier, J. Mitsugi, and G. Morabito, ‘‘The Internet of Things [Guest Editorial],’’ IEEE Wireless Commun., vol. 17, no. 6, pp. 8–9, Dec. 2010. [5] Z. Shelby and C. Bormann, 6LoWPAN: The Wireless Embedded Internet. New York, NY, USA: Wiley, 2011. [6] A. Zaslavsky, C. Perera, and D. Georgakopoulos, ‘‘Sensing as a service and big data,’’ CoRR, pp. 21–29, Jan. 2013. [7] E. Gons. (Dec. 6, 2012). The Internet of Things, accessed on May 12, 2013. [Online]. Available: http://www.flickr.com/photos/ wilgengebroed/8249565455/in/set-72157632169140208 [8] M. D. Hill, ‘‘What is scalability?’’ ACM SIGARCH Comput. Archit. News, vol. 18, no. 4, pp. 18–21, Dec. 1990. [9] M. N. Halgamuge, M. Zukerman, K. Ramamohanarao, and H. L. Vu, ‘‘An estimation of sensor energy consumption,’’ Prog. Electromagn. Res. B, vol. 12, pp. 259–295, Sep. 2009. [10] Q. M. Ashraf, M. H. Habaebi, and M. R. Islam, ‘‘Dynamic service arbitration for autonomic Internet of Things,’’ in Proc. 4th Int. Conf. Internet Appl., Protocols Services (NETAPPS), Cyberjaya, Malaysia, 2015, pp. 230–233. [11] L. B. Ruiz, J. M. Nogueira, and A. A. F. Loureiro, ‘‘MANNA: A management architecture for wireless sensor networks,’’ IEEE Commun. Mag., vol. 41, no. 2, pp. 116–125, Feb. 2003. [12] Q. M. Ashraf et al., ‘‘Energy monitoring prototype for Internet of Things: Preliminary results,’’ in Proc. IEEE 2nd World Forum Internet Things (WF-IoT), Milan, Italy, Dec. 2015, pp. 1–5. [13] J. O. Kephart and D. M. Chess, ‘‘The vision of autonomic computing,’’ Computer, vol. 36, no. 1, pp. 41–50, Jan. 2003. [14] Q. M. Ashraf, M. H. Habaebi, M. R. Islam, and M. T. Siddiqui, ‘‘Autonomic computing in Internet of Things for resource management,’’ J. Next Generat. Inf. Technol., vol. 6, no. 4, pp. 47–62, 2015. [15] Q. M. Ashraf and M. H. Habaebi, ‘‘Introducing autonomy in Internet of Things,’’ in Proc. 14th Int. Conf. Appl. Comput. Appl. Comput. Sci. (ACACOS), Kuala Lumpur, Malaysia, 2015, pp. 1–7. [16] M. B. Alaya, S. Matoussi, T. Monteil, and K. Drira, ‘‘Autonomic computing system for self-management of machine-to-machine networks,’’ in Proc. Int. Workshop Self-Aware Internet Things (Self-IoT), 2012, pp. 25–30. [17] C. Perera, A. Zaslavsky, P. Christen, M. Compton, and D. Georgakopoulos, ‘‘Context-aware sensor search, selection and ranking model for Internet of Things middleware,’’ in Proc. IEEE 14th Int. Conf. Mobile Data Manage. (MDM), Milan, Italy, Jun. 2013, pp. 314–322. [18] C. Perera, A. Zaslavsky, P. Christen, and D. Georgakopoulos, ‘‘CA4IOT: Context awareness for Internet of Things,’’ in Proc. IEEE Int. Conf. Green Comput. Commun. (GreenCom), Besançon, France, Nov. 2012, pp. 775–782. [19] Y. Yang and M. Cardei, ‘‘Adaptive energy efficient sensor scheduling for wireless sensor networks,’’ Optim. Lett., vol. 4, no. 3, pp. 359–369, Aug. 2010. [20] J. Deng, Y. S. Han, W. B. Heinzelman, and P. K. Varshney, ‘‘Scheduling sleeping nodes in high density cluster-based sensor networks,’’ Mobile Netw. Appl., vol. 10, no. 6, pp. 825–835, Dec. 2005. [21] J. Husney, ‘‘System and method for optimized initial configuration distribution for cyclic sleeping devices,’’ U.S. Patent 8 036 138 B2, Oct. 11, 2011. [22] S. Silas, K. Ezra, and R. B. Rajsingh, ‘‘A novel fault tolerant service selection framework for pervasive computing,’’ Human-centric Comput. Inf. Sci., vol. 2, no. 5, pp. 1–14, 2012. [23] S. Wang, Z. Zheng, Q. Sun, H. Zou, and F. Yang, ‘‘Cloud model for service selection,’’ in Proc. IEEE Conf. Comput. Commun. Workshops (INFOCOM WKSHPS), Shanghai, China, Apr. 2011, pp. 666–671. [24] J. Bao, Y. Ding, and H. Hu, ‘‘A new service selection algorithm in USPIOT,’’ in Proc. IEEE Int. Conf. Comput. Sci. Autom. Eng. (CSAE), Zhangjiajie, China, May 2012, pp. 22–26. 1320

[25] W. Li, Y. Zhong, X. Wang, and Y. Cao, ‘‘Resource virtualization and service selection in cloud logistics,’’ J. Netw. Comput. Appl., vol. 36, no. 6, pp. 1696–1704, Nov. 2013. [26] K. P. Yoon and C.-L. Hwang, Multiple Attribute Decision Making: An Introduction, vol. 104. Newbury Park, CA, USA: Sage, 1995. [27] G.-H. Tzeng and J.-J. Huang, Multiple Attribute Decision Making: Methods and Applications. Boca Raton, FL, USA: CRC Press, 2010. [28] F. Bari and V. Leung, ‘‘Multi-attribute network selection by iterative TOPSIS for heterogeneous wireless access,’’ in Proc. 4th IEEE Consum. Commun. Netw. Conf. (CCNC), Jan. 2007, pp. 808–812. [29] X. Hailun, W. Ning, and Z. Zhimin, ‘‘Neighbour peer selection scheme based on effective capacity for mobile peer-to-peer streaming,’’ China Commun., vol. 10, no. 5, pp. 89–98, May 2013. [30] Q. M. Ashraf, M. H. Habaebi, G. R. Sinniah, and J. Chebil, ‘‘Broadcast based registration technique for heterogeneous nodes in the IoT,’’ in Proc. Int. Conf. Control, Eng., Inf. Technol. (CEIT), Sousse, Tunisia, 2014, pp. 1–6.

QAZI MAMOON ASHRAF is currently pursuing the Ph.D. degree in computer engineering with the Department of Electrical and Computer Engineering, University Islam Antarabangsa, Malaysia. He was a Research Assistant with the Wireless Communication Division in MIMOS, Malaysia. He is also with the Digital Communications Laboratory, Telekom Malaysia Research and Development, where he is involved in Internet of Things (IoT) research. His research interests include IoT, autonomic computing, ubiquitous networks, and secure M2M communication.

MOHAMED HADI HABAEBI received the degree from the Civil Aviation and Meteorology High Institute, Libya, in 1991, the M.Sc. degree in electrical engineering from Universiti Teknologi Malaysia, in 1994, and the Ph.D. degree in computer and communication system engineering from University Putra Malaysia, in 2001. He is currently an Associate Professor and the Post Graduate Academic Advisor with the Department of Electrical and Computer Engineering, International Islamic University Malaysia, where he heads the research works on Internet of Things. He has supervised many Ph.D. and M.Sc. students, published more 120 articles and papers, and sits on the Editorial Boards of many international journals. He is actively publishing in M2M communication protocols, wireless sensor and actuator networks, cognitive radio, small antenna system and radio propagation, and wireless communications and network performance evaluation. He is an Active Member of the IEEE and an active reviewer to many international journals.

MD. RAFIQUL ISLAM (M’01) received the B.Sc. degree in electrical and electronic engineering from the Bangladesh University of Engineering and Technology, Dhaka, in 1987, and the M.Sc. and Ph.D. degrees in electrical engineering from the University of Technology Malaysia, in 1996 and 2000, respectively. He is currently a Professor with the Department of Electrical and Computer Engineering, Faculty of Engineering, International Islamic University Malaysia. He has supervised more than 20 Ph.D. and M.Sc. students and has published more than 150 research papers in international journals and conferences. His areas of research interest are wireless channel modeling, radio link design, RF propagation measurement and modeling, RF design, smart antennas and array antennas design, FSO propagation, and modeling. He is a Life Fellow of the Institute of Engineers Bangladesh.

VOLUME 4, 2016