3.1 Characteristics of Management Systems

  The way how MSs are deployed today is by establishing well-defined, central points of control, termed management platforms. As outlined in several publications (e.g. [3], [22]), platforms tend to become bottlenecks in large networks and a solution to this problem is to distribute their functionality (e.g., topology functions, event filtering mechanisms, logging facilities) in order to reduce the load on the central MS. This can be done by introducing MSs acting in the role of Mid-Level Managers (MLMs) that are located closer to the managed resources and execute management functions on behalf of other MSs. Their main purpose is to condense raw data stemming from the resources into meaningful management information that can then be used by another MS. This may also include caching of frequently accessed information like topology data of the underlying systems. These properties obviously reduce the amount of overall traffic that is sent through the network for management purposes. MLMs are an effective mechanism for structuring networks into domains and can be used for establishing management hierarchies. An extension of this concept is to delegate (and withdraw) management functionality to MLMs at runtime. This delegation may either be initiated by the MLM itself (pull model) or by another MS (push model). The interworking between MSs of different service providers as described in section 1 can be seen as a special case of interactions between different MLMs. Consequently, the above mentioned principles apply also to independent MSs operated by different authorities.

As motivated in section 2, further complexity is introduced by the heterogeneity of the deployed management architectures. MGs are a convenient mechanism for achieving interoperability between heterogeneous MSs. They are obviously located on the boundaries of different (architectural) management domains and act in the role of an MLM w.r.t. other MSs. As MGs are aware of the architectural specifics of the underlying resources, it therefore makes sense to provide them with the same management functionality as MLMs. Such an enhanced MG could e.g., condense several SNMP-traps into one CORBA-event instead of translating every trap into one event and leaving the filtering of these events to the MS one level above in the service provider hierarchy. This implies that providers need an instrumentation for these systems in order to present a unified view on their management data to customers; the amount of accessible information may be specified in a service contract.