Another group of candidates for application control in distributed systems are toolkits for supervising parallel programs[VS95,KS99,GES98,MCC$^$95]. They allow fine grained access to the application data structures for debugging and monitoring purposes. Their upcoming has been driven by the need to increase the degree of interaction as opposed to batch systems that offer only limited access to the programs at run-time. For these reasons, programs that are controlled by a steering system are instrumented with code that comes with such toolkits. Although this fine grained access allows the definition of new commands that affect specific parts of the application, more efficiently than with process based batch systems, it is difficult to cover a variety of different programs under the same scope of control. The user interface of the toolkit allows interaction with one executable image, the toolkit server on each node. This implies that tasks have to execute in the scope of exactly this process. Interaction and on-line visualisation of data are major issues, but nevertheless no effort has been made to offer different data communication policies. As a result traditional event-stream oriented systems tend not to scale well[ES98]. In high-energy physics environments local area and wide area networking applications are used together. It is for example desirable that an operator retrieves information about the progress of an experiment that might even be installed in a laboratory on a different continent. In that case application steering systems for parallel programs would exhibit low performance due to exactly their fine grained communication patterns. But also in large computing clusters, special solutions have to be found if continuous monitoring of all application nodes is wanted, since the tools do not allow for buffering or threshold driven delivery of gathered application data.