Comparing Process and Event-Based Software Performance Simulation

Predicting the performance of complex software systems most commonly relies on simulation. For this, executable computer models (simulators) are developed that resemble the performance of real systems. The way in which the system behaviour is captured by simulators is strongly influenced by the so-called worldview employed by the simulation developer. The two most common world-views are processand event-oriented simulation. The choice for a world-view does also affect the performance and scalability of the resulting simulator. This has been addressed by a few studies comparing the world-views in terms of performance, which, however, lack of representativeness. Specifically, none of them is based on a simulator actually being used for performance predictions. Therefore, we perform such a comparison in the context of the Palladio Component Model (PCM), which has been used successfully in a number of case studies, thus emphasizing its representativeness. We compare the process-oriented PCM simulator SimuCom to its event-oriented counterpart EventSim, where the latter one has been developed in this thesis. We i) identify factors potentially influencing simulation performance, ii) rank these factors, and iii) assess the simulator’s performance and scalability based on the most influential factors. A validation suggests that EventSim is semantically equivalent to SimuCom. For most PCM models, EventSim outperforms SimuCom due to the overhead induced by the process-orientation. Both simulators scale linear in terms of simulation duration when increasing the model complexity; the simulation duration in SimuCom, however, increases at a rate up to 50 times the rate of EventSim. Limits in scalability have been observed only with SimuCom, but most of them can not be attributed to the worldview used.