T. Vaskivuo (Finland)
Software visualisation, performance analysis, correlation, dimensionality reduction
In this paper we investigate methods for analysis and visualisation of run-time behaviour of software in embedded systems. Methods for measuring run-time software behaviour are limited in the context of embedded products. Precise measurements require instrumented hardware and software, and the arrangements may be expensive and difficult to be set up, especially in a small product line. To compensate this, the measurement methods presented in this paper are as independent of the precise product platform as possible. We use purely software-based sampler that extracts statistical information about software execution. In this limited context, we investigate the use of different classification methods and correlation analysis in creating effective interpretations of run-time behaviour. Presented methods can be used in analysis and tracking of any repeating activity in the run-time behaviour of software during product evolution. In validating the analysis methods, we use a simulation environment in creating data that has known origin. Additionally, we show results in analysing data that is measured from an ARM-based Symbian device. Results indicate that the presented set of methods provides an inexpensive and effective way to analyse complex software behaviour in software design of embedded systems.
Important Links:
Go Back
