Performance modeling of computing processes using reconfigurable differential stochastic Petri nets 23 PERFORMANCE MODELING OF COMPUTING PROCESSES USING RECONFIGURABLE DIFFERENTIAL STOCHASTIC PETRI NETS

A natural modeling framework for many complex systems, such as communication or computer systems and networks, is provided through discrete event systems and, in particular, generalized stochastic Petri nets (GSPN) models [3]. However, factors such as huge traffic volumes, increasingly complex operating rules, and performance requirements make such models highly impractical. From an analytical standpoint, traditional models from classical queuing theory fail to capture new features such as complicated traffic source behavior or blocking phenomena. An alternative modeling paradigm for the purpose of analysis and simulation is based on Stochastic Fluid Models (SFM). The SFM paradigm allows the aggregation of multiple events into a single event associated with a “significant change” in the system dynamics. Among the formalisms of SFM that are used, the fluid stochastic Petri nets (FSPN) [9] and hybrid stochastic Petri nets (HSPN) [1, 8] are popular. To make design issues and analysis procedures more transparent with negative-continuous values, we tried to deviate as little as possible from the concepts and the nets of FSPN and HSPN. Thus, we propose our extension of differential Petri nets (GDPN) [5], which we call Generalized Differential Stochastic Petri Net (GDSPN), and that is able to represent the behavior of computing processes in a common model. The features of GDSPN accept the negative-continuous place capacity, negative real values for continuous place marking and tokendependent arc cardinalities that permit to generalize the concept of GDPN, FSPN and HSPN.