Projekt

Performance models may provide reliable answers concerning the quality of service of computer systems. Design decisions in the model building process represent a crucial step, which also affect the applicability of available evaluation algorithms. Analytical techniques are usually much faster than simulations, but there exist many system variations and growing models become complicated soon. But what must a model look like to get trustworthy answers, what are the key parts of a certain computer system which must be modeled and where are the limits of particular models? This is the starting point for our project. We want to provide a framework which shows a practicable way of how to start with modeling a certain infrastructure, which might exhibit some performance problems. It is a main goal to investigate the influence of the different components in the system on the accuracy of the modeling results. Besides, we want to compare different modeling approaches in view of their outcomes. In the end, this will demonstrate the limits of simple analytical or simulative models. The objects being modeled are part of the infrastructure of our cooperation partner, the University of Applied Sciences in Kempten: on the one hand a web-server and on the other hand a distributed web-service, i.e., an online portal which consists of a web-server, an application-server and a database-server. Because these systems are real productive systems, we are convinced to meet the requirements of many other teams using comparable infrastructures in a commercial way. We get real data from these systems which deliver the information to parameterize and validate the models. In addition, we evaluate, respectively develop, some tools for gathering the required numbers about the computers. At the moment we investigate a straightforward solution for measuring delays on the network with only one PC without further time synchronisation.