Stochastic performance models are widely used to analyse the performance and reliability of
systems that involve the flow and processing of customers and resources. However, model
formulation and parameterisation are traditionally manual and thus expensive, intrusive and
error-prone.
This thesis illustrates the feasibility of automated performance model construction from high-precision
location tracking data. In particular, we present a methodology based on a four-stage
data processing pipeline which automatically constructs Coloured Generalised Stochastic Petri
Net (CGSPN) performance models from an input dataset consisting of raw location tracking
traces. The output performance model can be visualised using PIPE2, the platform independent
Petri Net editor. The developed methodology can be applied to customer-processing systems
which support multiple customers classes and can capture the initial and inter-routing probability
of the customer flow of the underlying system. Furthermore, it detects any presence-based
synchronisation conditions that may be inherent in the underlying system and the presence of
service cycles. Service time distributions, one for each customer class, of each service area in the
system and travelling time distributions between pairs of service areas are also characterised.
PEPERCORN, the tool that implements the developed methodology, is also presented.
In addition to the latter, this thesis presents LocTrackJINQS, the extensible, location-aware
Queueing Network simulator. LocTrackJINQS was developed to support location-based
research. It has the ability to simulate a user-specified Queueing Network and while simulation
progresses, it generates and outputs location tracking data – associated with the movement of
the customers in the network – in a trace file.
Our methodology is evaluated through six case studies. These case studies use synthetic location
tracking data generated by LocTrackJINQS. The obtained results suggest that the
methodology can infer the abstract structure of the system – specified in terms of the locations
and service radii of the system’s service areas (max error 0.320 m and 0.277 m respectively) and
customer flow – and approximate its service time delays well. In fact, the maximum relative
entropy value that was obtained between the simulated and inferred service time distributions
is 0.324 nats. Furthermore, whenever synchronisation between service areas takes place, the
simulated synchronisation conditions are successfully inferred.