Understanding the dynamic properties of a large distributed software system is most challenging. Knowledge of a system's dynamics is a prerequisite for the design of viable solutions for both coding and testing. Taking the perspective of a maintainer of the system, we have focused on how execution trace data generated from object interaction during execution can be used in maintenance.

A practical method for improving a system (by increased understanding) is a contribution of this research. The result increases our understanding of how information based on execution tracing can be derived and used. To support the method we developed a toolbox consisting of CORBA-based systems comprising trace collection, information building and visualisation. In short, trace collection is performed by observing the messages sent over the CORBA layer, thus no instrumentation of source code is needed. Information building consists of pattern matching and statistical analysis. Visualisation is performed with the Spotfire.net^TM tool.

The method has been examined during a case study with software optirnising operation of a network for GSM. Trace data were collected during both testing, and operation at a customer site. By feeding back generated information to the design organisation several positive results have been noted. For example, performance tuning has become a natural activity amongst designers, as the system behaviour is better understood from information building on both testing and normal operation. In the actual case we were able to suggest nine improvements based on unexpected use and non-use of the systems parts.

The conclusion is that we have found a practically relevant area, where only little information has been published. For the future, we are keen on generalising the findings with more studies and to increase the technical sophistication and theoretical foundation of the toolbox.