Policy based management framework
To realise the management of complex clusters with a high number of nodes it is necessary to automatically come to a decision on the meaning of sensor data on different levels of abstraction. Though near process data principally has other imtent on the system behaviour as i.e. netowrk information and therefore has different impact on the bevahiour to solve or prevent failures. Just this correlation of data can lead the preventive actions to cause further errors. So the process move to a node which will break down soon doesn't make sense at all.
Classic cluster management is based on administrator knowledge regarding the system infrastrukture and typical error indicators. Generally they use their own experience - historic data - to make the proper analysis. This workpackage aims in particular on replicating and automating these general behaviour and analytical methods of administrators. For this a maximun in flexibility for possible patterns must be ensured from a simple information system up to specific process calls. It has to provide a intuitive method to define analysis rules and policies to cope with the system complexity.
The workpackage follow the approach of "Knowledge Engineering" which especially means:
- Analysis of typical administator knowledge and behaviour considering different situation criteria (Interviews)
- Abstraction and transfer of these informations into a machine readable and situation dependant description (Structuring)
- Development of the logical System based on the semantic database (Coding)
- Transfer of logical rules, policies and knowledge onto the system (Specifiing)