Page 53 - Data Science Algorithms in a Week

Using Deep Learning to Configure Parallel Distributed Discrete-Event Simulators 37

Figure 7: Combined Speedup chart for BTW, BTB, and TW for different number of processors (nodes)
– A global node is a separate cluster. A local node is a node from a specific cluster. Therefore Global 3
and Local 3 means 3 separate clusters and each one with 3 computers (in total 9 nodes).

Characterization of Software Complexity

Measuring simulation algorithm complexity is challenging. Researchers have
proposed measures that categorized complexity by measures such as number of codes
lines, code internal structures, and interfaces. Shao and Wang (2003) and Misra (2006)
examined software complexity with the perspective of software being a product of the
human creative process. As such, they explored complexity measures based on cognitive
weights, which takes into account the complexity of cognitive and psychological
components of software. In this paradigm, cognitive weights represent the effort and
relative time required to comprehend a software piece. The approach suggests that
software complexity is directly propositional to the complexity of understanding the
information contained in it. We have selected this measure because is the most
recognized in the literature.
Using cognitive weights of basic control structures to measure complexity addresses
the cognitive and architectural aspects of software complexity. Basic fundamental logic
blocks of software constructs such as conditional if-then statements, method calls, for-
loops, etc. are assigned a weight value. Table 2 shows the cognitive weights of each type
of basic software control structure (BCS).