Challenges and Experiences in Transitioning Management Instrumentation from Command-Oriented to Model-Driven - Advances in Network-Embedded Management and Applications - page 6

characteristics that should be considered when designing a model-driven
architecture based on a command-oriented system.
3.1 Defining the Instrumentation Model
The instrumentation model may be defined using either of two distinctly
different methods. It may be derived from legacy source code or it may be
explicitly constructed based on domain knowledge.
Derivation from Source Code
In an effort to minimize the impedance mismatch between the imposed model
and the implicit model, deriving the instrumentation model from the
management instrumentation implementation is often considered by
designers. Since the CLI management interface often contains the most
complete implementation it is frequently seen as the canonical source for
model derivation; however, this task can be wrought with difficulty.
Achieving this objective requires correct and complete interpretation of
implementation source code sufficient to extract instrumentation
configuration data elements, operational data elements, and services. Model
element extraction alone is not enough to meet objectives, as the interpretation
of the semantic relationship of features, data, and services is also required in
the modeled system. Without a perfect interpretation, gaps and model
generation errors will require exhaustive human interaction and domain
experience to correct. Moreover, it should be carefully considered if the
implicit instrumentation model in-fact meets the needs of the target model-
driven system.
Designing from Domain Knowledge
Designing an instrumentation model relies heavily on the designer's
knowledge of the instrumentation domain space. They must understand the
configuration data, operational data and services offered by the
instrumentation and the relationships between them. If modeling an existing
system, it should be accepted that regardless of modeling choices, there is
going to be some level of unavoidable impedance mismatch with legacy
systems; however, when modeling new instrumentation in the model-driven
system, this is not the case. Designers creating models for brand new
instrumentation can ensure models have a 100% match to data and services.
When constructing model architectures, designers should avoid sacrificing
model extensibility by designs that are too rigidly tied to legacy structures.