Model for performance analysis of distributed stream processing applications

Abstract

Nowadays, a lot of data is produced every second and it needs to be processed immediately. Processing such unbounded streams of data is often applied in a distributed environment in order to achieve high throughput. There is a challenge to predict the performance-related characteristics of such applications. Knowledge of these properties is essential for decisions about the amount of needed computational resources, how the computations should be spread in the distributed environment, etc. In this paper, we propose a model to represent such streaming applications with the respect to their performance related properties.We present a conversion of the model to Colored Petri Nets (CPNs) which is used for performance analysis of the original application. The behavior of the proposed model and its conversion to the CPNs is validated through experiments. Our prediction was able to achieve nearly 100% precise maximum delays of real stream processing applications.