Applying Information Fusion To Monitor Service Of Complex System

With the rapid development of cloud computing, mobile Internet and wearable devices, the enterprise server has a increasing scale and the client device has miniaturization and individuation features. The properties of massive data, millions of machines, heterogeneous node and complex interaction in a large distributed system present enormous challenges. In order to meet these challenges, monitor service should collect real-time data, such as resource utilization, task schedule, quality of service, etc. to detect and alert anomalies in the system. Then it improves the reliability of the system. Therefore, a monitor service plays an indispensable role in a complex environment. However, the traditional monitor service has two drawbacks. First, current services didn’t give a clear definition of internal structure and relationship about monitor data. Instead, they used a form of name-value pair in the log or database. Second, they dealt with different dimensions of data simply and isolatedly and didn’t cover the aspects in data noise removing, feature extraction, data fusion and decision making. As a result, users have to pay extra cost to find anomalies from complex multi-dimension data.To solve these problems, this paper proposes a complex monitor service with information fusion. Along with the name, value, type, measurement, time and location aspects of monitor data, a three-level monitor framework is designed, which includes resource layer, logical layer and service layer. Meanwhile, data fusion algorithm joins between the layers so that the low-level data can be fused into high-level data and generates intuitive decision data which reflects the status of the whole system. Based on publish/subscribe middleware, we further realize monitor service. The execution process is defined as follows. Ⅰ)Map the node, frame and component in the middleware into the host, process and thread in the monitor data using ontology modeling. Ⅱ)Collect the resource data referring host, process and network through a plug-in on Windows or Linux system. Ⅲ)Fuse the resource data into logical data which reflects the health condition of a node by means of threshold, Three-Sigma and fuzzy Bayesian. Ⅳ)Aggregate resource and logical data of all nodes to the central node. Ⅴ)The central node adopts the ontology fusion and generates the service data which reflects the load balance status in the system. Ⅵ)The user can access the visual charts through a browser. Meanwhile, the alert e-mail or message will be sent if any anomaly happens.In order to verify the reliability of monitor service, a experiment with data distribution application on four PCs is designed and implemented. In addition, a comparison with availability and accuracy between different fusion algorithms is accomplished. Experiment results show that the monitor service can get the resource utilization of each node correctly and detect the abnormalities timely in heterogeneous, complex interactive environment.