مدیریت جریان کاری مبتنی بر فیلتر بلوم قادر به تضمین QoS در شبکه های حسگر بی سیم

As a popular service composition technology, workflow has been successfully used in wireless sensor networks (WSNs) to compose a set of atomic services for service-oriented WSN applications. However, in a resource-constrained WSN, the sensed data is usually inaccurate or even missing, and this affects the normal execution of atomic services and may result in the non-guaranteed workflow QoS. Because the implementation of workflows in WSNs is usually hierarchical, effective workflow management in a WSN should consider both aspects of atomic services and sensor nodes, though this has largely been overlooked in existing research. Hence, a dynamic QoS-oriented, effective and efficient hierarchical workflow management mechanism is necessary. In this paper, we propose a Bloom filter-based hierarchical workflow management model, which coordinates both the atomic service level and the node level for guaranteed workflow QoS. Through constructing the service-level counting Bloom filter (CBF) to maintain the set of normal atomic services, and constructing the node-level Bloom filter (BF) to maintain the set of attribute strings of the current working nodes, an effective and efficient QoS degradation locating can be realized. Furthermore, the corresponding adaptation mechanism for guaranteed QoS is also developed. The case study and experimental evaluations demonstrate the capability of the proposed approach in WSNs.

As the development of wireless technology, wireless sensor networks (WSNs) have been widely deployed in various applications. However, each WSN application usually has its own pre-designed implementation structure and this makes the interoperability among WSN applications difficult. Fortunately, service-oriented architecture (SOA) ( Erl, 2005) provides the characteristic of supporting collaborations among distributed autonomous applications in an open dynamic environment. Hence, many attempts have been made to bring SOA into WSNs for effective realization of cooperative, interoperable and loose-coupled applications ( Leguay et al., 2008, Gu et al., 2005 and Delicato et al., 2003). Due to the efficiency and the practicability, workflow has become a popular service composition technology, through which a sequence of atomic services can be composed together to satisfy the specified functionality. However, practical SOA applications usually contain many atomic services with compatible functionalities. This provides the basis for the service composition to pay attention to the QoS requirements ( Cardoso et al., 2004 and Chen et al., 2011). Take an example of temperature control application, in which the current temperature determines whether or not to turn the air conditioner on. If a user wants to achieve the temperature control functionality with two QoS requirements on cost≤100cost≤100 and View the MathML sourcedeviation≤1°C respectively, an abstract workflow can be built and composed of a set of abstract atomic services like View the MathML sourcestart→datagathering→datareasoning→actionexecution→end Turn MathJax on Let us assume that two concrete atomic services exist to implement the data reasoning service: • S 1 with QoS attribute definition (cost=60, deviation≤5°Cdeviation≤5°C) and • S 2 with QoS attribute definition (cost=90, View the MathML sourcedeviation≤0.5°C). If S1 is chosen, the user's expected functional can be achieved, though the deviation of S1 (5 °C) is beyond the user's expectation (1 °C). This renders the composed workflow suboptimal for this user. Hence, a large amount of effort has been devoted to satisfy different QoS requirements simultaneously ( Zeng et al., 2004 and Yu et al., 2007). However, due to factors such as the changes in environment, the QoS attributes of atomic services can change or even become unavailable ( Rinderle et al., 2004) during the execution of corresponding workflows. Consequently, these changes may increase the risk of a composed service with poor QoS, which will further render the corresponding workflows incapable of meeting users' expected QoS. Accordingly, QoS-oriented dynamic workflow adaptation is necessary as this is a key issue in workflow management. Workflow adaptation means that the workflow should adapt itself to the changing environment in order to satisfy both the functional and the QoS requirements. Methods for implementing the workflow adaptation can be divided into two categories: ( Zeng et al., 2004, Ardagna and Pernici, 2007, Alrifai and Risse, 2009 and Cardellini et al., 2009) recomposing and replacing. Recomposing re-selects the sequence of appropriate atomic services to compose a new workflow with guaranteed QoS, and can achieve the global adaptation by considering the QoS of the whole workflow. While through replacing, abnormal atomic services will be replaced by those with the compatible functionalities. The replacing method belongs to the local adaptation, as it only considers the sanctification of the faulty service's functionality (e.g. input and output), while fails to guarantee the whole workflow QoS for lack of a global view. Unlike traditional SOA, workflow management in a WSN is much more flexible. Due to the data-centric characteristic and the strong data gathering capability of a WSN, atomic services are inevitably redundant. Hence, the simple adoption of SOA into WSN may bring issues such as the waste of application resources or inefficient service management. In order to alleviate this problem, Tong et al. (2011) proposed a reasoning-based context-aware workflow management model (Recow) for WSNs. In this model, a rule-based reasoning module extracts semantic information so that the lower-level sensor data will have a loose-coupled connection with the upper-level logic process. By deploying these semantic atomic services whose inputs are characterized with semantic information, Recow can build a flexible workflow by taking advantage of its loose-coupled connection with the sensor level. However, as the implementation of workflows in a WSN is usually hierarchical, faulty atomic services or sensor nodes will affect the correct execution of workflows. Effective workflow management in WSNs should consider both the aspects of atomic services and the aspects of sensor nodes. Accordingly, this calls for an effective and efficient hierarchical workflow management mechanism. More specifically, the following two challenges should be addressed: • As WSNs are limited in energy and computation resources, the gathered sensor data can be inaccurate because of energy exhaustion, device fault or environmental obstruction. However, as WSN applications are data-centric, the inaccuracy in the lower-level sensed data will affect the correct execution of upper-level atomic services, thus further deteriorate the corresponding workflow QoS. This forms the bottom-to-up dynamical characteristic of WSN applications. • In the hierarchical WSN workflow management, the lower level hides the implementation details for the upper level. While most WSN applications have a requirement of low latency, once the workflow QoS deteriorates to an unacceptable threshold, the hierarchical management should locate and fix the problem promptly. Due to these challenges, an efficient management mechanism is in high demand. As the Bloom filter is efficient in representing the huge number of elements, it can be utilized to improve the efficiency of workflow management in a WSN. More specifically, hierarchical Bloom filters can be constructed to represent atomic services and sensor nodes. However, the effective interaction among different levels of Bloom filters remains an open and challenging issue, which is also the research objective of this paper. Hence, we will firstly exploit the Bloom filter at each level, and then through flexible interaction between the hierarchical Bloom filters, an efficient hierarchical workflow management in WSNs can be achieved. Accordingly, this paper will further focus on the following research issues: • A hierarchical workflow management model. We will propose the hierarchical workflow management model, which coordinates both the atomic service level and the node level for guaranteed workflow QoS so that it is suitable for resource-constrained WSNs. • Bloom filter-based QoS-oriented workflow adaptation mechanism. Firstly, dynamic frequency is utilized in the QoS monitoring for effective detection of the QoS degradation. Then, if the QoS degradation occurs, Bloom filters on the service level will interact with Bloom filters on the node level to locate the source of the QoS degradation. Finally, workflow adaptation can be activated to guarantee the workflow QoS again. The rest of this paper is organized as follows. Section 2 reviews the related work, followed by the proposed hierarchical workflow management model in Section 3. Section 4 presents the workflow adaptation mechanism for both the service level and the node level. Section 5 provides experiments that illustrate the benefits of the proposed approach. Finally, Section 6 concludes this paper.

In this paper, we propose a Bloom filter-based hierarchical QoS-oriented workflow adaptation mechanism for WSNs, which tries to achieve workflow adaptation on both the atomic service level and the node level. Then, we developed a workflow management model with four modules: the QoS monitoring, the planner, the service-level adaptation and the node-level adaptation. In this model, we constructed the service-level CBF to maintain the set of current normal atomic services, and constructed the node-level BF to maintain the set of attribute strings of the current working nodes. When the workflow QoS degraded, and with the help of the service-level CBF and the node-level BF, we realized an effective and efficient locating of QoS degradation. The planner module then coordinates the service-level and node-level adaptation to guarantee workflow QoS again. Compared to existing work on workflow management, the proposed approach is suitable for resource-constrained WSNs. Results from experiments indicate that the proposed approach is much more efficient in space and time compared with the simple linked list-implemented approach, and it also achieves a greater success ratio to guarantee workflow QoS in WSNs compared with a traditional workflow management approach. Our future research will continue to focus on node-level management, particularly which action should be taken to improve the corresponding atomic service QoS after identifying abnormal sensor nodes.