Delivering Serverless Experience on Kubernetes: Beyond Web Applications
summary
Evaluates Knative on Kubernetes for two non-web workloads and builds prototypes: a run box proof-of-concept that uses a persistent volume plus a deployment scaled 0↔1, integrates with Jupyter kernels via kubectl exec, supports on-demand vertical resizing, and preserves interpreter state; and a hybrid GPU/CPU inference setup that implements a custom scheduler and a pool of warm pods to avoid 20 seconds cold start. Demonstrates the prototypes work end-to-end, surfaces limitations in native GPU scheduling and cold-start handling, and proposes lightweight shared-pod and vertical-resize mechanisms.
AI-written summary · about
chapters (6)
- 0:00Start
- 4:57Serverless = Distraction-Free (Examples)
- 6:52Serverless Compute Platform (SCP)
- 9:27SCP: Function as a Service (FaaS)
- 22:14SCP for Interactive Computing
- 34:03Our Architecture
related talks
23mServerless with Knative
20mKnative: A Kubernetes Framework to manage Serverless Workloads
52mServerless with Knative
31mBeyond Traditional DevOps: Dive into Serverless with K8s
34mModern Serverless Workloads with KNative and Kubernetes
21mServerless computing using Knative Open Source
26mUnleashing the Power of Serverless Computing in Kubernetes with Knati...
57mServerless with Knative