Failure to respond to the users requests within an acceptable amount of time means that system is unavailable to its users. Therefore the prediction of systems response time for different intensity of users requests, esp. identifying requests loads causing performance degradation becomes an important issue in the design of a dependable Internet system (incl. e-commerce systems).

A method for modeling application server environment by the use of queueing networks has been proposed. The physical components of the environment are represented as service centres which form a separable queueing network. Model parameters, i.e. average service times and average visit counts can be experimentally measured in a test environment. Model output are response times for different request classes as well as other parameters describing systems performance. Tools supporting the measurements and model calculations have also been developed. Experiments carried out on real system show that the model predicts real systems performance with satisfactory accuracy, esp. maximum throughput of the system.