Most decision trees can be straightforwardly represented as a series of convergent processes performed in sequence, each with the purport, ``If it is broken, then fix it''. A typical convergent process consists of a test and perhaps a configuration change action. If the test succeeds, no action is performed; else an appropriate action is taken in order to ensure that the test succeeds at a later time.
The processes in most decision trees are not convergent in this sense. Sometimes it takes a bit of worthwhile effort to fit an existing decision tree to the convergent process model. For example, one might replace the decision tree above by three convergent processes:
This convergent representation has advantages over the original decision tree representation. If all three of these processes report success, ftp is running as desired. It no longer matters which process is tried first, as the diagnostic procedures will not take action upon a problem unless the problem has been observed directly and a change is necessary.
In our study we have been unable to generate a network decision tree that cannot be represented as a set of convergent scripts. Network decision trees seldom exhibit more complexity than can be represented straightforwardly in this form. Even when they do, such trees can be expressed in this simpler form by encapsulating more complex processes inside larger ``convergent'' processes that replace several decisions.