We present the Ka-admin project that addresses the problem of collecting, visualizing and feeding back any grid information, trace or snapshot, compliant to an XML-like model. Real use includes performance analysis of parallel applications and cluster administration. Ka-admin includes a generic “filter” module that processes monitored data independently of what they represent. Filters can remove, aggregate and transform data or pass them to external applications. The end user is responsible for activating the filters from within an interactive graphical interface. This allows him to focus on important information. We also present a “scatter/gather” module that allows efficient collection and distribution of data and commands in a large cluster. Early work on “MPI/threads” applications and system monitoring tools proved that the combination of both modules matches the objective of a scalable visualization of large data sets.