Scalar: A Distributed Scalability Analysis Framework

Abstract

Analyzing the scalability and quality of service of large scale distributed systems, such as cloud based services, requires a systematic benchmarking framework that is at least as scalable to sufficiently stress the system under test. This paper summarizes Scalar, our distributed, extensible load testing tool that can generate high request volumes using multiple coordinated nodes. It has support for communication and synchronization between user threads, and built-in node monitoring to detect resource bottlenecks in the benchmark framework deployment itself. Furthermore, it offers highly scalable results analysis that exploits data locality and characterizes the overall system scalability in terms of the Universal Scalability Law.