Placement of measurement points  

The following section describes how measurement points can be inserted into the source code of an application without requiring manual intervention of the application developer. This is done for both the monitoring of user transactions and the monitoring of subtransactions.

General model of a user transaction  [r]

2#2

Monitoring of user transactions:  As already mentioned, the monitoring of user transactions is of great importance. A general model of a user transaction is depicted in figure . It starts with a user interaction (UI) initiated by the user (e.g., pressing a GUI button or the ENTER-key) followed by some kind of activity. At the end of the user transaction, the result is presented to the user.

At development time, the application developer must identify the interactions that start and end the user transactions to be monitored (laid down in the SLA) and must uniquely name these user transactions. As in the case of component based application development no source code access is available for the application developer, this can only take place during component customization. This means, every component providing user interactions must be (manually) instrumented by the component developer to allow this kind of identification. As only a small number of components provide interactions and as GUI components achieve a particular high degree of reuse, this additional effort can easily be tolerated.

Thus, every input component must provide a means to declare each of its user interactions as the beginning of an arbitrary user transaction. To distinguish different user transactions, a means to uniquely name transactions is further required. Whenever such an interaction takes place, the respective component must inform a measurement agent about the beginning of a new instance of a user transaction. Analogously, a way to declare the end of a user transaction is needed. The correlation of the two measurements can be done automatically as described in section .