An instrumentation and visualization technique for performance analysis of high-performance industrial embedded applications

Most performance analysis toolsets developed for parallel systems are oriented to measuring and analyzing high-performance scientific applications. Recently, these toolsets have evolved from managing message-passing scientific applications to managing data parallel scientific applications. In the same way, we consider that new techniques must be incorporated in performance analysis toolsets, in order to adapt them to applications other than scientific ones, for example to the high-performance industrial embedded applications in our case. To carry out this adaptation, in this paper, we propose two techniques referred to as model driven instrumentation (MDI) and model driven visualization (MDV). The basic idea of these techniques is, first, to build a behavioral model of the application to be analyzed and later, use this model to drive the instrumentation and visualization of the application. The use of these techniques is explained by means of a simple synthetic example.