Scalability Evaluation of ExplorViz with the Universal Scalability Law

Modern distributed stream processing systems play an important role in cloud computing systems and Big Data applications. To cope with varying intensity of user load, an important characteristic that often is required for such systems is scalability. The Universal Scalability Law is a performance model to describe the scalability of universal systems. In this work, we examine to what extend the Universal Scalability Law can be integrated with the methodology of the Theodolite benchmarking framework for cloud-native stream processing systems. Theodolite assesses scalability based on service level objectives (SLOs). We find that the Universal Scalability Law can be used to make the execution of benchmarks more efficient with regard to the total execution time and to quantify the scalability based on the benchmark results. However, we find that due to the measurement method used by the Theodolite framework the interpretability of the model may be influenced by the underlying SLOs. We apply our extended version of Theodolite to benchmark the scalability of the software visualization and comprehension framework ExplorViz, which visualizes monitored applications using dynamic analysis. ExplorViz comprises a microservice-based architecture that uses the stream processing framework Kafka Streams. Our results show that some of the ExplorViz microservices scale linearly, but that the system scalability is mainly bounded by one microservice