Applications

  Several research projects have come up with utilizations of such models. One of the applications is the so called root cause analysis. It helps to find a common (root) cause of problems or faults detected at distinct places within an environment. It may be applied to network components reporting error conditions as well as to services where, end users detect problems. The reason for the actual need of such root cause analysis is that error conditions or problem reports brought to administrators or management systems, are just descriptions of symptoms. To be able to derive their causes, further knowledge about the dependencies among them is necessary. [#!grus98b!#], [#!hsv99!#] and [#!kapa97!#] explain this subject in detail.

Similar dependency models are needed when determining availability requirements on sub-services (looking from a top down perspective) respectively for the calculation of service availability from the availability of underlying services (bottom up), as described in [#!kais99p!#].

Knowledge of dependencies between systems may be of further use for the prediction of impacts on other systems due to management operations. This is of particular interest in typical maintenance scenarios, where a server has to be shut down temporarily: It is essential to know, resp. to simulate the effects on other systems beforehand. Further investigations of advantages can be found in [#!pemt95!#] and [#!clem94!#]. A common result of their and others' examinations is that--assuming models do already exist--great benefits can be achieved for management tasks. For our purposes, following major advantages for the practical utilization of environmental models can be resumed; they:

More applications of environmental models emerge if the algorithm used for their generation allows--like the one presented in section --frequent iteration of the modeling process in certain time intervals. This enables the analysis of changes of dependencies in the managed environment during that time. This is, e.g., useful for fault prediction, because significant changes in the overall system behavior are detected through emerging or disappearing dependencies. This often reflects errors that are already present in currently unused parts of a service which may later (under different usage conditions) effect its usability. The detected changes may also be used to point out forbidden actions or disallowed use of services. This is helpful especially for intrusion detection and to recognize service misuse.