Using
Design of Experiments to Automatically Support
the Calibration of Performance Models
Beatriz
Salvo-Garau, Nunzio-Nicolo Savino-Vazquez, and Ramon
Puigjaner
Dept. of Mathematics and Computer Science
Universitat de les Illes Balears
Crtra. de Valldemossa, km. 7.5
E-07071 Palma de Mallorca, Spain
E-mail:
beatriz@ipc4.uib.es
savino@ipc4.uib.es
dmirpt0@clust.uib.es
Simulation can be
successfully applied to design performance-responsible
software once a pool of models of the target hardware and
the services-oriented software have been tested and
validated. In this paper we present SD/CAM (Statistical
and Deductive Calibration Method) and its implementation
in MOCA (Model Calibration Assistant) as a method for
conducting the calibration of performance models. It can
be applied either during the design and implementation of
lower layer performance models or the
validation of higher layer performance models. SD/CAM is
based on considering as inputs the different values on
which a performance model can be parameterised, and the
results obtained by simulating the model as observed
variables. Analysis and Design of Experiments can be
therefore applied in this modelling context.
SD/CAM is composed of
several steps:
- Definition of the
performance model essential characteristics to be
statistically validated.
- Collection of
real measures from the system whose model is
being validated.
- Matching model
specifications with measures, in order to
establish the statistical model of the system.
- Model simulation,
in order to obtain a pool of relations about
results and characteristics of the model.
- Evaluation and results analysis to derive
conclusions about the significance of the model.
<BACK TO
PROGRAM SCHEDULE> <HOME>
|