Large communication networks typically consist of multiple interconnected subnetworks and span a number of different organizations. For the purpose of managing such networks, each organization is generally certain to have its own network management system which reflects different management policies. Although each network management system is operated by different organization, there are many cases that it should need sharing of management data and information of others. Therefore, in the perspective of information sharing, interoperability between network management systems is demanded as a mandatory requirement for muti-domain network management. In order to support interoperability of network management systems, it is important to provide them with an ability of locating, selecting or accessing other's management data necessary for managing their own network efficiently.
In this paper, we propose an architectural model to provide interoperability between network management systems and import operations for importing and selecting necessary data out of multiple information sources as a primary means for offering interoperability. The import operations provide management applications with functionality of expressing constraint and preference over target data.
In case that a data import request is aimed at a number of data stored at local database systems of different network management systems, the request inevitably becomes a global transaction that targets multiple databases. A major issue in scheduling global transaction is how to ensure global serializability. To assure global serializability, we propose Global Ticket Method which uses global tickets. The global ticket is a logical timestamp whose value is stored as a data item at management data broker that is a core component in the proposed architectural model. In order for Global Ticket Method(GTM) to ensure global serializability, global serializability validation algorithms which are executed on management site are proposed. The global serializability validation algorithms prevent global subtransactions from being serialized in any different order with assigned global tickets at participating local database systems. The major advantage of GTM is to reduce communication overhead by preventing global restarts caused by both out-of-order ticket operations and ticket conflicts, which are major problems of Optimistic Ticket Method(OTM) proposed by Georgakopoulos. However, GTM could lead to low degree of concurrency because the serialization order of global transactions is determined in a top-down fashion by global tickets in advance.
Accordingly, in order to investigate trade-offs between GTM and OTM, we have conducted performance comparison on basis of a fair foundation. Our experimental results showed that GTM outperforms OTM in cases that short timeout values are given. In OTM, it is found that, in cases that the relatively short timeout values are given, the global deadlocks could be detected early. This leads to the effect of increasing the number of global transactions competing for active status and in turn increases the probability of ticket conflicts. However, in case that the timeout value relatively becomes long, our results demonstrated that OTM outperforms GTM. The reason is found as follows. In OTM, in case that a long timeout value is given, global deadlock detections are delayed and the global transactions in deadlock state are blocked until the deadlocks are resolved. This has the effect of reducing the number of global transactions in active state, and in turn reduces the probability of ticket conflicts. However, in case of GTM, the number of global transactions waiting for the completion of preceding global transactions is increasing as the timeout value increases. Moreover, if the preceding global transactions are blocked with global deadlocks, the successive global transactions can suffer restarts due to the delayed deadlock detection consequent upon the timeout expiration of successive global transactions.
