Scheduling Hardware-Accelerated Cloud Functions

Abstract This paper presents a Function-as-a-Service (FaaS) approach for deploying managed cloud functions onto heterogeneous infrastructures. Current FaaS systems, such as AWS Lambda, allow domain-specific functionality, AI, HPC and image processing, to be deployed in the while abstracting users from infrastructure platform concerns. Existing approaches, however, use single type of resource configuration execute all function requests. In this paper, we present novel that allows effectively executed across compute resources, including hardware accelerators GPUs FPGAs. We implement scheduling tailor selection each request, taking into account performance cost way, our makes different processor types quantities (e.g. 2 CPU cores), uniquely suited handle workload, potentially providing improved at reduced cost. validate three application domains: machine learning, bio-informatics, physics, target with combined computational capacity 24 FPGAs 12 cores. Compared traditional FaaS, achieves improvement non-uniform traffic up 8.9 times, maintaining objectives.