A Logical Architecture for Active Network Management

Salvatore Gaglio, Luca Gatani, Giuseppe Lo Re, Alfonso Urso
2006 Journal of Network and Systems Management  
This paper focuses on improving network management by exploiting the potential of "doing" of the Active Networks technology, together with the potential of "planning," which is typical of the artificial intelligent systems. We propose a distributed multiagent architecture for Active Network management, which exploits the dynamic reasoning capabilities of the Situation Calculus in order to emulate the reactive behavior of a human expert to fault situations. The information related to network
more » ... ts is generated by programmable sensors deployed across the network. A logical entity collects this information, in order to merge it with general domain knowledge, with a view to identifying the root causes of faults, and to deciding on reparative actions. The logical inference system has been devised to carry out automated isolation, diagnosis, and even repair of network anomalies, thus enhancing the reliability, performance, and security of the network. Experimental results illustrate the Reasoner capability of correctly recognizing fault situations and undertaking management actions. purposes of management, particularly for configuration setting, fault diagnosis, and performance evaluation, but, as the size of networks increases, it becomes more and more difficult to extract the right information from them. Conventional network management facilities have been provided by network vendors in ad hoc ways, which rely heavily on human effort. However, as networked installations become larger, more complex, and more heterogeneous, manual network management is no longer able to cope and conventional approaches are therefore no longer effective. The complexity of such systems raises the cost of network management and requires the use of automated standardized tools that can be used in complex scenarios, across a broad variety of product types. The main purposes of network management are to maintain a network in healthy operational condition, to monitor the network status, and to control the network so as to maximize its efficiency. Current network management systems are typically designed according to a centralized paradigm, where a central station (manager) collects, aggregates and processes data retrieved from physically distributed devices (agents). Widely deployed standards, such as the Simple Network Management Protocol (SNMP) [1] of the TCP/IP protocol suite, or the Common Management Information Protocol (CMIP) [2] of the OSI reference model, are designed according to this strict centralized model. However, the centralized approach is characterized by a low degree of flexibility and re-configurability, suffering from severe inefficiencies and scalability limitations: the process of data collection and analysis typically involves massive transfers of data causing considerable strain on network throughput, as well as processing bottlenecks at the central entity. Taken together, these problems suggest that the distribution of management intelligence would offer a rational approach to overcoming the limitations of the centralized approach. The Internet Engineering Task Force (IETF) has therefore proposed an approach, known as RMON (remote monitoring) [1], which introduces a degree of decentralization. A key aspect is that the collection of management information should be supported in a timely way, so that the system can react to performance problems. In addition, monitoring traffic should ideally have a minimal impact on the managed network. Several distributed architectures for network management have been proposed to relieve the load from the central management station and to distribute control tasks by means of the Active Networks technology. Active Networks [3, 4] introduce network dynamic programming and allow an easy deployment of "ad hoc" solutions in the active nodes on behalf of contingent management tasks. Given simple management tasks (such, for instance, multiple failure traps, data merging, automatic backup-link activation), management architectures based on Active Networks make easy to deploy a distributed strategy. Nevertheless, for more complex tasks, it is still required the human intervention because only experts, who know the network complexity, can understand a high-level management goal and can plan a sequence of intermediate steps, in order to reach the objective. In a traditional network management environment, if a user asks to know the causes of
doi:10.1007/s10922-005-9012-7 fatcat:l4uzg4tnprfb3mzngvn324gmqy