Each of the outlined approaches tackles a specific problem. Batch systems allow timed invocation and provide load balancing algorithms. They are however difficult to extend and their interaction granularity is coarse. Application steering systems offer direct interaction with the programs through code instrumentation, but are limited to control and monitoring of single applications. The required access to the source code of the applications that have to be controlled is a severe limitation in domains that have to deal with heterogeneous systems and legacy software. Agent systems offer a flexible and clean way of interacting with a variety of different tasks. The limited ability to be controlled individually has however to be taken into consideration. Furthermore techniques to reduce the overhead of code and state management have to be addressed to allow their deployment in highly efficient application domains. Each of the paradigms alone can probably not satisfy the needs of controlling applications in high performance distributed data acquisition systems. Still, the agent pattern[SD98] seems to be an ideal mapping to our problems and could in combination with the qualities of the first two approaches provide a solution.