Exploiting software repositories

A new extension to these approaches is described in [#!kakc2000!#]. Additionally to the analysis of application configurations it makes use of information stored in software installation repositories of operation systems to obtain intra-system relationships.

In every case the result of this step is one dependency model reflecting the current stage of the environment. Step 3-iii indicates the reiteration of the previous two steps, which leads to a sequence of models being generated over time. The appropriate duration of the time period between the models also depends on the modeling approach and application. Looking at configuration files might be needed only about once a day, while our approach allows (if needed) to reflect the actual service utilization in arbitrary periods of time, e.g., for every hour.

For the interaction points between this phase and the next it is crucial to define a common model exchange format. Several of the formats mentioned in section (like MOF, etc.) suite this purpose. In [#!enke01!#] we developed an XML/RDF-based format specially suited for dependency description in distributed environments. However, for the purpose of this paper the actual representation is of minor interest.

Once the model is generated the last phase starts. Management tools use the generated models directly (step 4-i) or calculate and use comparisons of sequentially generated models (step 4-ii), e.g., to point out emerging differences (which indicate changes in the modeled environment) to the administrator or to generate alarms to management platforms if dependencies to certain important services change.

Special management cases may also require to derive an abstract service model from the environmental one. This helps to find unusual cases of unexpected dependencies either indicating inconsistencies in the model or errors in the network or service design.