At the Chair of Computer Architecture and Parallel Systems, Jianfeng Gu, Mohak Chadha, Anshul Jindal, and Michael Gerndt are working on the development of more fine-grained GPU resource sharing mechanism for Serverless Computing to further enable heterogeneous resource scheduling of Function Delivery Network (FDN) project.
The credits for Google Cloud will help to (i) implement the fine-grained MPS-based sharing and isolation mechanism of the GPU resources, (ii) design a custom resource controller to extend the serverless platform (e.g. Openfaas) to support auto-scaling and address cold-start problem, (iii) optimize the pod scheduling strategy which dynamically allocates functions to heterogeneous nodes and (iv) benchmark the overall performance of deep learning based inference tasks.
Google Cloud Research Credits worth $5000 awarded to Jianfeng Gu for his proposal on “FaaSShare: Fine-grained GPU Resource Sharing for FaaS”