Run-time monitoring of system performance: A goal-oriented and system architecture simulation approach

The performance of a software system has been a major concern of the information technology services industry since its inception in the early 1960´s. Up to now, simulation has been a key technique for gaining an understanding of the performance characteristics of the system during development time, while software tools, possibly together with manual operations, have been used to monitor some performance characteristics of the system during run-time. Simulation models, by and large, have not been tied closely (not easily traceable) to the requirements; neither have they been tied to the run-time monitoring. In the past, simulation models were constructed and used in design and simply ignored during run-time. In this paper, we propose a goal-oriented framework (implemented as a softgoal interdependency graph to record goals, development alternatives, design rational and linkage to functional requirements) and a system architecture simulation approach to realize and monitor the run-time performance characteristics of the system. This approach treats performance requirements as softgoals: a system architecture is constructed that meets the softgoals, a simulation model is constructed and experiments analyzed to consider varying workloads, resource consumptions, and run-time capacities. A good-enough system architecture is then used to implement the system and help monitor if the run-time behavior of the system indeed matches the performance requirements. Alternative actions, to solve performance miss-matches, can be evaluated using the simulation model. The approach is illustrated using a previously underperforming government tax-payer system.