Multicloud Deployment of AI Workflows Using FaaS and Storage Services
Manju Ramesh, Dheeraj Chahal, Rekha Singhal · 2023
Multicloud deployments are getting traction for benefits such as agility, avoiding vendor lock-in, resilience, etc. Use of Function-as-a-Service (FaaS) platforms is emerging as a preferred choice for deploying AI workflows. These platforms from different cloud service providers (CSPs) have unique specifications and cost models. The onus is on a user to find a cost-performance optimal mapping of its application workflows to FaaS and its associated services in a multicloud deployment setting. In this work, we have presented an empirical study of multicloud deployment of AI inference-workflows using FaaS and cloud storage services. Our evaluation shows that the cost for executing AI workflow reduces by 83% using an optimal combination of CSPs over a naive deployment. Further, we also propose and evaluate analytical models for estimating the cost of deployment in multicloud using FaaS and storage services.