17-04-2012, 12:47 PM
Model-driven generative development of measurement software
modeldriven.pdf (Size: 915.04 KB / Downloads: 49)
Introduction
Metrics offer a practical approach [25,48] to evaluate nonfunctional
properties of artifacts resulting frommodel-driven
engineering (MDE) development processes. Ledeczi et al.
[25] showed that a use of the models is design-time diagnosability
analysis to determine sensor coverage, size of ambiguity
groups for various fault scenarios, timeliness of diagnosis
results in the onboard system, and other relevant domainspecific
metrics. More recently, a similar point of view is
expressed by Schmidt et al. [48]: in the context of enterprise
distributed real-time and embedded (DRE) systems, our system
execution modeling (SEM) tools help developers, systems
engineers, and end users discover, measure, and rectify
integration and performance problems early in the system’s
life cycle.
How to generate measurement software?
The measurement approach presented in this paper is
intended to be applied to models of a model-driven engineering
(MDE) software development, i.e., applied tomodels
fully described by a metamodel, which is later called domain
metamodel. Our intuition was [34] that it is possible to generate
measurement software from an abstract and declarative
specification of metrics. This abstract level for metrics
can be defined by a metamodel as well. In other words, the
measurement software is generated from a model of metric
specifications.
The metric specification metamodel
The MDM approach consists in specifying metrics as an
instance of ametric specificationmetamodel.The metamodel
is a definition of the metric concepts. It grounds the complete
generation of the measurement software from an instance of
themetamodel, called a metric specificationmodel. The concepts
of the metric specification metamodel can be applied
to any domain models. For instance, the same concept can
be instantiated as a metric for implementation models, realtime
models, software architecture models or requirements
models. In that sense, our metric specification metamodel
is domain-independent [36]. It has been built following a
bottom-up approach: from existing metrics to concepts of
the metamodel. Hence, all concepts are instantiated several
times.
Evaluation: application cases
In this section, we present the application of the MDM
approach into different contexts. For each application case,
we present: (1) the motivation for measuring, (2) the domain
metamodel, and (3) the metrics and the corresponding specifications
with respect to the metric specification metamodel.
This section presents the facts that will be further analyzed
in Sect. 5.
Conclusion
Our model-driven measurement (MDM) approach allows
modelers to automatically add measurement capabilities to a
domain-specific modeling language. Whatever the domain,
whatever the maturity of the product during the development
life cycle, it allows the effective measurement thanks to a
complete generation of the measurement software from an
abstract and declarative specification of metrics. Our prototype
is able to generate an integrated and full-fledge domainspecific
measurement software from this specification.