Collaborative Research: SHF: Small: Exploiting Performance Correlations for Accurate and Low-cost Performance Testing for Serverless Computing

University Of Delaware, Newark DE

Investigators

Abstract

As organizations increasingly port their applications to cloud services instead of using local resources, they face the challenge of determining how to use cloud resources and maintain good interactive experiences for their application users while reducing their cloud usage costs. Serverless computing greatly simplifies the deployment of large-scale cloud applications by automatically provisioning and managing the servers and removing this burden from cloud developers. However, cloud developers still need to determine the most cost-effective cloud-resource allocations for their deployments and debug performance issues in their serverless applications. This research focuses on providing cloud-application developers with accurate performance knowledge by addressing the lack of performance testing tools to accurately determine the performance of serverless cloud applications. The success of this research can reduce the deployment costs and improve the performance satisfaction for organizations, such as education, research, and health institutions, that utilize serverless clouds. This research specifically addresses the scalability and workflow challenges that serverless computing sets forth to accurate performance testing through the design of novel performance-testing frameworks that will utilize correlation-aware, non-parametric statistical tools, including clustered-block bootstrap and regression-based bootstrap. It will provide three accurate, low-cost, and automated serverless performance-testing frameworks, specifically for: 1) unit testing for a single serverless application's performance by exploiting the correlation within the application's performance testing data from simultaneous invocations (i.e., intra-application correlation); 2) integration testing for the performance correlation between two serverless applications (i.e., inter-application correlation); and 3) integration testing for the overall performance of a workflow of serverless applications by exploiting both intra- and inter-application correlations. The insights and techniques developed in this project will also deepen the research community's understanding of the performance characteristics of serverless computing and on how to effectively analyze serverless performance using modern statistical tools. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.