Birds of a Feather: Do You Really Need on Premises Serverless ?
summary
Prototype demonstrates running OpenWhisk on Kubernetes across x86, Power and ARM nodes to serve data‑science pipelines and video analytics, using CouchDB for JSON results, Kafka for event distribution and TensorFlow checkpoints for model state. Builds multi‑arch containers with Docker manifest multi‑arch images (x86, ppc64le, ARM) and deploys to clusters including Jetson Nano edge GPUs so training/inference stays local. Works as a proof‑of‑concept; on‑prem requires nontrivial integration and developer effort but improves GPU utilization and reduces data movement.
AI-written summary · about
chapters (17)
- 0:00Start
- 1:58What is serverless ?
- 2:38History
- 3:10Main use cases
- 3:38My first use case
- 5:07OpenWhisk flow
- 5:46OpenWhisk architecture
- 7:11OpenWhisk git projects
- 8:57Serverless and data science
- 10:57Function and data science
- 11:11Jupyter nb flow process
- 11:41TF implementation Save and restore a model
- 12:02Target architecture
- 14:40Docker for multiple architectures
- 15:32Scheduler customisation
- 26:10Is all actions should be hardware agnostic ?
- 27:44You need on premise serverless if ...
related talks
34mApache OpenWhisk: Building a Production-Ready Serverless Stack on/for Kubernetes
27mCloud Native, Event Drive, Serverless, Microservices Framework - OpenWhisk
56mApache OpenWhisk
55mApache OpenWhisk: How it Works and How You Can Benefit
05mServerless architectures built on an open source platform
38mHow to Build a Distributed Serverless Polyglot Micro Services IoT Platform
06mKeynote: Cloud Native Architectures with an Open Source, Event Driven, Serverless Platform
15mServerless Apps with Node.JS in OpenWhisk and Adobe I/O Runtime