Performance Comparison of Function-as-a-Service Triggers

Abstract

Cloud computing paved the way for how servers are handled and maintained, and recent developments in cloud computing have established a new paradigm shift to wards serverless computing. Through a subset called Function-as-a-Service (FaaS), most operational concerns are abstracted away and allows developers to focus en tirely on the code (i.e. functions) to be executed. FaaS functions are triggered by events (called triggers) and there are many types of triggers offered by each provider. This thesis studied the latency of three trigger types through a trace-based approach. The three triggers were HTTP triggers, storage triggers, and queue triggers. To fur ther contrast previous work, the comparisons were also made across two providers, Amazon Web Services (AWS) and Microsoft Azure. Focus was also put on dis cussions justifying the comparison between two largely different providers and on the reproducibility of the study. The HTTP trigger performed the best for both providers, the Queue trigger second-best for AWS and third-best for Azure, and the Storage trigger third-best for AWS and second-best for Azure. In terms of providers, both performed relatively similarly in terms of mean delay but Microsoft Azure had significantly more extreme outliers compared to Amazon Web Services. In conclu sion, the study performed in this thesis found that the choice of service and provider can greatly affect a system’s performance and can, in extension, affect the usage of cloud services.