System management for distributed data acquisition systems is demanding in terms of the capabilities to include diverse software and hardware. It has been shown that the requirements can be met by providing a toolkit that avoids hardcoding of policies and applies the agent pattern for the tasks that are subject to control. Although the solution to the given problem is quite simple, we do not see commercial support for this approach. Instead we find tools for computing cluster management from major IT companies, targeted at office networks that are extensible only with vendor provided modules. On the other side there exist science-born approaches for academic computing environments that are not flexible enough. Concerning the problems posed by mixed environments or large distributed real-time systems, we do not find any approapriate solutions. Assumptions do not belong in scientific work, but the trend of increased distribution and the pushing forward of embedded systems into the commercial world will soon make it necessary to provide flexible tools for application steering. Choices of concrete technologies for implementation, e.g. CORBA for communication or Java as implementation platform should be postponed as long as possible. What counts is the simplicity of the approach and the possibility of tailoring the tool to the given configuration.