Implications for Running AI Applications on Serverless Platforms

Artificial Intelligence (AI) has garnered extensive attention due to its robust computational capabilities, efficient data handling, and superior accuracy. However, the complexity inherent in AI applications poses significant managerial and cost-related challenges for AI practitioners. There is a burgeoning trend towards leveraging serverless technologies to support AI applications, owing to their ease of deployment, automatic scalability, and cost-effectiveness. In this paper, we present a systematic performance evaluation of serverless applications with varying characteristics and structures on both X86 and ARM platforms. We utilize two of the most popular open-source serverless platforms, Knative and OpenFaaS, in our assessment. The insights gained from our evaluation are invaluable for the effective deployment of applications across X86 and ARM platforms. We are confident that the implications drawn from our experiments will not only inspire AI engineers to design more effective principles for AI applications on serverless platforms but also guide them in efficiently executing AI applications within such environments.