Abstract: In this paper we investigate the use of stochastic models for analysing service-oriented systems. We propose an iterative hybrid approach using system measurements, testbed observations as well as formal models to derive a quantitative model of service-based systems that allows us to evaluate the effectiveness of the restart method in such systems. In cases where one is fortunate enough as to have access to a real system for measurements the obtained data often is lacking statistical significance or knowledge of the system is not sufficient to explain the data. A testbed may then be preferable as it allows for long experiment series and provides full control of the system's configuration. In order to provide meaningful data the testbed must be equipped with fault-injection using a suitable fault-model and an appropriate load model. We fit phase-type distributions to the data obtained from the testbed in order to represent the observed data in a model that can be used e.g. as a service process in a queueing model of our service-oriented system. The queueing model may be used to analyse different restart policies, buffer size or service disciplines. Results from the model can be fed into the testbed and provide it with better fault and load models thus closing the modelling loop.
Keywords: Fault model
Note: Proceedings of the Fourth International Workshop on the Practical Application of Stochastic Modelling (PASM 2009)
This page generated by bibTOhtml on Do 21. Sep 01:00:03 CEST 2017