Page 277 - Data Science Algorithms in a Week

258 Oloruntomi Joledo, Edgar Gutierrez and Hatim Bukhari

simulation, changes are represented as separate events to capture logical and sequential
behaviors. An event occurs instantaneously (such as the press of a button or failure of a
device) to cause transitions from one discrete state to another. A simulation model
consists of a set of rules (such as equations, flowcharts, state machines, cellular automata)
that define the future state of a system given its present state (Borshchev and Filippov,
2004).
A simulation can also be classified in terms of model structure. Sulistio, Yeo and
Buyya (2004) proposed a taxonomy encompassing different approaches. The presence of
time is irrelevant in the operation and execution of a static simulation model (e.g., Monte
Carlo models). For the case of a dynamic model, in order to build a correct representation
of the system, simulated time is of importance to model structure and operation (e.g.,
queuing or conveyor).
Dynamic systems can be classified as either continuous or discrete. In continuous
systems, the values of model state variables change continuously over simulated time. In
the event that the state variables only change instantaneously at discrete points in time
(such as arrival and service times), the model is said to be discrete in nature. Discrete
models can be time-stepped or event-stepped (or event-driven). In discrete-event models,
the state is discretized and "jumps" in time and the steps (time-step) used is constant.
State transitions are synchronized by the clock i.e., system state is updated at preset times
in time-stepped while it is updated asynchronously at important moments in the system
lifecycle in event-driven systems.
Deterministic and probabilistic (or stochastic) properties refer to the predictability of
behavior. Deterministic models are made up of fixed input values with no internal
randomness given the same output for same corresponding input. Hence, the same set of
inputs produces the same of output(s). In probabilistic models however, some input
variables are random, describable by probability distributions (e.g., Poisson and Gamma
distributions for arrival time and service times). Several runs of stochastic models are
needed to estimate system response with the minimum variance.
The structure of a system determines its behavior over time. Ecommerce system is a
complex, interactive and stochastic system that deals with various people, infrastructure,
technology and trust. In addition, factors like uncertainty, competition and demand
defines its economic landscape. These markets are non-linear, experiencing explosive
growth and continuous change. Developing representative models comprise of detailing
stakeholders and pertaining underlying processes. Decision makers must consider these
factors when analyzing the system and procuring optimal strategies to assess model
viability.