Scalar: Systematic Scalability Analysis with the Universal Scalability Law

摘要

Analyzing the scalability and quality of service of large scale distributed systems requires a highly scalable benchmarking framework with built-in communication and synchronisation functionality, which are features that are lacking in current load generation tools. This paper documents Scalar, our distributed, extensible scalability analysis tool that can generate high request volumes using multiple communicating, coordinated nodes. We show how Scalar offers analytics capabilities that support the Universal Scalability Law. We illustrate Scalar on an electronic payment case study, and find that the framework supports complex work flows and is able to characterize and give predictive insights into the quality of service and relative capacity of the system under test in function of the user load.