A cloud-based execution framework for program analysis

Abstract

Program analysis is a popular method to determine properties about program behavior, such as execution times and potential security vulnerabilities. One of the biggest challenges faced by almost every form of program analysis is scalability. One way to address scalability issues is to distribute the analysis across multiple machines. However, this is not an easy task; designing a distribution framework that is capable of supporting multiple types of program analysis requires careful thought and consideration. This paper presents the cloud-based execution framework that we built for performing distributed analysis of Java bytecode programs. We describe the design decisions that allow this framework to be generic enough to support multiple types of analysis but remain efficient at the same time. We also present a simple, static work partitioning algorithm that we have found to work well in practice and provide benchmarks to show its efficiency.