I was a native mobile and B2B software engineer (self taught). I am having to ramp my edge computing skills. Here is what has helped me so far .... /1
Balena has a very easy workflow. You can set up and provision and use many devices, configure your SSH key, push code to devices via GitHub. Lots of supported interesting applications. Let’s you provision a Pi and automatically get it on Wi-Fi /2
Really good starting point, if you can do all of the above in Balena then you have the basics. After that I started looking at @openfaas because it has a faasd version and a lot of great documentation on how to run serverless on a Pi with minimal Kubernetes /3
In talking with @rhatr on the @eve_edge meet up on Friday - it seems a lot of people want to avoid Kubernetes for now at the edge because of the complexity and lift, so serverless kinds of approaches like @alexellisuk is innovating might become more popular /4
Next I plan to run through this demo on NVIDIA EGX stack + Kubernetes running DeepStream on Jetson to get a feel for Cloud-Native Edge AI https://developer.nvidia.com/blog/deploying-ai-apps-with-egx-on-jetson-xavier-nx-microservers/ /5
Objective is to get the basics of AI on the edge, learning about Helm and Kubernetes and how to stand up an edge AI device with cloud native basics. /6
Currently I am playing with @eve_edge. I think we have many architectural approaches emerging. Eve thinks the edge will look like Android on an embedded hyper visor. It was easy to setup and run on a Pi, recommend getting involved in projects like this https://github.com/lf-edge/eve/blob/master/README.md /7
recommend reviewing and having knowledge of the following open source projects and groups:
- ACRN hypervisor
- LFEdge Group
- LFAI&Data Group
- PARSEC (CNCF) for edge security
- Zephyr
- EdgeX
/8
- ACRN hypervisor
- LFEdge Group
- LFAI&Data Group
- PARSEC (CNCF) for edge security
- Zephyr
- EdgeX
/8
Next, look at Baetyl https://baetyl.readthedocs.io/en/latest/develop/install.html.
I plan on spending a lot of time playing with this because it takes a Kubernetes approach. I want to see how easy it is to set up and run but the architecture interests me. Contributed by Baidu. It uses K3S from Rancher /10
I plan on spending a lot of time playing with this because it takes a Kubernetes approach. I want to see how easy it is to set up and run but the architecture interests me. Contributed by Baidu. It uses K3S from Rancher /10
Edge needs lightweight Kubernetes. Recommend investigating K3S from Rancher as a lightweight edge k8
Also look at Canonical microk8 https://microk8s.io/
https://rancher.com/docs/k3s/latest/en/ /11
Also look at Canonical microk8 https://microk8s.io/
https://rancher.com/docs/k3s/latest/en/ /11
I think having experience configuring and standing up edge clusters will be a useful skill, a lot of@interesting hardware projects to help people do this like @turingpi and @gumstix and @BitScopeDesigns
http://turingpi.com /12
http://turingpi.com /12
Hearing a lot of good things about @Azure DevOps to deploy K8s to edge being used to solve real problems. Plan to investigate that myself shortly, there was a great talk by Paul DeCarlo on how to do this on NVIDIA https://developer.nvidia.com/gtc/2020/video/s21118-vid /13
On the microcontroller side highly recommend @EdgeImpulse , they make it really simple to learn about AI on constrained devices with minimal nonsense or need for PHD /14
Further reading, if you want to position yourself for maximum impact seems programming languages will be Go, Rust and eventually WebAssembly.
Follow @beriberikix and his work on WebAssembly and AI on embedded devices. WebAssembly feels like it could enormous eventually /15
Follow @beriberikix and his work on WebAssembly and AI on embedded devices. WebAssembly feels like it could enormous eventually /15
Read the top five blogs on the @bytecodeallies news to understand. New API for NN and growing usage of WebAssembly adoption at the edge already https://bytecodealliance.org/articles/ may be a few years off but this feels huge. /16