Title:  Towards An Execution Optimization in Serverless Functions Through A Spatial-Temporal Orchestration Mechanism for Context Sharing in Highly Concurrent Conditions

Committee:

Dr. Vijay Madisetti, ECE, Chair, Advisor

Dr. Raheem Beyah, ECE

Dr. Chuanyi Ji, ECE

Dr. Arijit Raychowdhury, ECE

Dr. Ling Liu, CoC

Abstract: Serverless computing has become a popular model for deploying scalable applications in a fully cloud-managed environment where developers only pay for the execution and not idle time. A key enabling service, Function-as-a-Serverless (FaaS), is the primary execution paradigm offered by cloud providers. FaaS offerings, such as AWS Lambda, Azure Functions, and Google Cloud Functions, emerged to provide an environment to deploy function-based applications with minimal intervention from the developer. Although this new paradigm aims to create the right environment to improve developer velocity, the capacity to scale complex multi-tenant functions processing events in a highly concurrent scenario is inefficient. In this thesis, we examined the performance challenges of allocating and orchestrating these functions. Furthermore, to assess the performance, we present an asynchronous multiprocessing profiler tool to access the function execution model and analyze provider efficiency when executing functions at multiple concurrency levels. To address this inefficiency, we introduce a spatial-temporal and cross-region execution orchestrator capable of handling a stream of requests to functions in a scalable and efficient manner. Moreover, this orchestrator aims to improve the allocation of resources to compute these applications, optimizing cloud resources while minimizing the latency of requests across many users. The aim is to transform the FaaS offering from a point-to-point service model where clients interact to a many-to-one execution model while dynamically optimizing the resource usage of these functions over time.