Interoperable Hardware Acceleration for Serverless
summary
vxl is a framework that decouples function calls from hardware implementations so a single binary runs on Jetson Nano/Xavier, Intel NCS, Nvidia T4 and custom NPUs without recompilation; plugins map a static image-classify API to vendor-shipped shared objects, and vm guests call accelerators via virtio/vsock or API remoting. Integrated with Firecracker and Kata/OpenFaaS, measured end-to-end overhead is typically <5%, and the project is open source with C/Rust/Python bindings.
AI-written summary · about
chapters (4)
related talks
39mDelivering Serverless Experience on Kubernetes: Beyond Web Applications
20mFirecracker: Lightweight Virtualization for Serverless Applications
15mSabre: Hardware-Accelerated Snapshot Compression for Serverless MicroVMs
24mLightweight VMs for serverless and containers
31mBirds of a Feather: Do You Really Need on Premises Serverless ?
34mKubeFlow’s Serverless Component: 10x Faster, a 1/10 of the Effort
46mMonolith to Serverless - Can Modern Cloud-Native Java Runtimes Do It All
35mMoonshot Spark: serverless with GraalVM
