F3: An FPGA-accelerated FaaS Framework

FPGAs provide a programmable, energy-efficient, and computeintensive acceleration substrate; thus, on the one hand, they offer a compelling solution for optimizing serverless workloads in cloud environments. On the other hand, FPGAs also introduce significant challenges that directly contradict the serverless model in the cloud, including their low-level and complex programming APIs, lack of virtualization and isolation mechanisms, high reconfiguration and communication overheads, and absence of orchestration mechanisms.

To this end, we present F3, the first system that enables efficient and secure use of FPGAs to accelerate serverless functions. F3 allows serverless functions to easily offload their tasks to FPGAs without worrying about low-level device management. The system automatically deploys and invokes FPGA-accelerated functions while providing isolation, high throughput, and low latency. To achieve these design goals, F3 simplifies the complexity of FPGA initialization by a high-level hardware-agnostic API, enables function isolation with unikernel-based FPGA virtualization, reduces cold starts and communication overheads with a per-node FPGA resource manager, and maximizes FPGA utilization with an FPGA-aware orchestrator.

We implement an open-source prototype of F3 based on a practical serverless framework, which consists of OpenFaaS, Kubernetes, and containerd, following the CRI/OCI industry standards for cloud orchestration engines. Our evaluation demonstrates significant latency reduction and throughput improvements for real-world application workloads, microbenchmarks, and Azure production traces.