Performance Analysis of Computing Servers using Stochastic Petri Nets and Markov Automata

Generalised Stochastic Petri Nets (GSPNs) are a widely used modeling formalism in the field of performance and dependability analysis. Their semantics and analysis is restricted to “well-defined”, i.e., confusion-free, nets. Recently, a new GSPN semantics has been defined that covers confused nets and for confusion-free nets is equivalent to the existing GSPN semantics. The key is the usage of a non-deterministic extension of CTMCs. A simple GSPN semantics results, but the question remains what kind of quantitative properties can be obtained from such expressive models. To that end, this paper studies several performance aspects of a GSPN that models a server system providing computing services so as to host the applications of diverse customers (“infrastructure as a service”). Employing this model with different parameter settings, we perform various analyses using the MaMa tool chain that supports the new GSPN semantics. We analyse the sensitivity of the GSPN model w.r.t. its major parameters –processing failure and machine suspension probabilities– by exploiting the native support of non-determinism. The case study shows that a wide range of performance metrics can still be obtained using the new semantics, albeit at the prize of requiring more resources (in particular, computation time).