Mohammad Shahrad

Serverless computing is an emerging paradigm in cloud computing that revolutionizes application development. It introduces unique features such as a pay-per-use pricing model and the ability to scale down to zero. By shifting the responsibility for resource provisioning and administrative tasks to cloud providers, it frees developers from the complexities of resource management, allowing them to focus solely on code logic. However, despite the numerous advantages of serverless computing, there is a potential for decreased efficiency particularly when used in conjunction with other cloud services, raising environmental concerns. Recent data from leading cloud providers highlight compute servers, storage servers, and network equipment as significant sources of carbon emissions in cloud computing. Thus, there is a pressing need for enhanced resource efficiency, particularly in the context of serverless computing.This thesis addresses the resource and cost efficiency of today's serverless functions by minimizing the allocation footprint for serverless functions. The UnFaaSener framework is introduced, leveraging the underutilized capacity of non-serverless hosts to execute serverless functions. By using available underutilized resources within a user's non-serverless computational resources, the framework eliminates the need for cloud providers to allocate additional resources, thus mitigating potential resource overallocation for the serverless functions.Furthermore, this thesis explores the integration of the WebAssembly (Wasm) execution environment as the isolation mechanism for serverless functions. When serverless functions have the potential to be compiled into Wasm modules, unlike traditional containers, Wasm can enable the efficient packing of a larger number of modules onto the same isolation environment, maximizing resource efficiency through mitigating the considerable memory footprint associated with containers. The proposed techniques contribute to creating more efficient serverless applications for users, offering potential improvements in cost-effectiveness and performance.

View record