A Vision Turned into a Production Reality All the Way to the Edge!

March 1, 2023
Posted by: Nati Shalom
Category: 5G, AWS, Cloudify, Edge Computing, Hybrid Cloud, Intel, MWC, Network Orchestration, TensorFlow

Cloudify @ MWC 2023

During MWC 2019, we first announced project Spire. Project Spire was the codename for our Cloudify edge strategy. It was designed primarily for handling orchestration at extreme scale (10k sites, x millions of devices). The design was based on a unique distributed orchestration architecture as could be seen in the diagram below.

_{Figure 1 Cloudify Project Spire – Next Generation Orchestration: Multi Domain, Sites, Clouds, Hyper-Scale}

Fast forward, Cloudify is being used now to manage one of the largest 5G carrier production networks and we’re continuously introducing performance optimization to support this scale. We are now able to improve our discovery performance by an order of magnitude, in addition to adding the ability to cope with continuous drifts of the underlying edge environment.

In addition to that, we’ll continue to introduce new innovations in this area.

End to End Network, Edge, and Application Orchestration

This year at MWC 2023, we’re demonstrating an end-to-end service orchestration of network, edge, and application combined led by AWS TSB and together with six partners such as Mavinier who’s delivering the 5G services, Intel and Dell for providing a far edge Kubernetes (EKS-A) infrastructure that is optimized for low latency, and Proto who’s delivering 3D hologram streaming application at the edge. This integration also introduces support for ETSI standards through the new AWS API for network management.

On a personal note, I must admit that when I first wrote my piece on the Public Cloud Effect on Telco I didn’t realize how quickly that vision will become the new norm. Today, we can’t even think of any network service without seeing it tightly integrated with public cloud infrastructure at its core.

Similarly, when I first started to introduce the Agile-First Approach I didn’t imagine how much it will simplify the way we develop and manage such complex use cases. This project is a perfect demonstration of that. Having 6 vendors working remotely with only a few engineers working part-time from different time zones couldn’t be done without the layer of consistency, rich ecosystem, and built-in services that AWS provides.

_{Figure 2 End-to-End Network, Edge, and Application automation with AWS TSB, Mavenir, Dell, Intel, and Proto}

Creating a Self-managed and Adaptive Machine Learning (ML) Inference Network

Additionally, we’re demonstrating our support for optimizing Machine Learning (ML) Inference (the process of running live data points into an ML model to calculate an output, such as a single numerical score) all the way to the edge. In this specific case, we’re using Jupyter Notebooks: software, standards, and services for interactive computing, and TensorFlow for a software library of machine learning and artificial intelligence which is specifically optimized for training and inference of deep neural networks. Lastly, we’ll use ResNet-50 a 50-layer convolutional neural network (a class of artificial neural networks, most commonly applied to analyze visual imagery).

Machine Learning systems tend also to be highly distributed and data intensive. In this context, Cloudify uses the ML Ops layer to simplify the management of such a network. Furthermore, Cloudify’s ability to continuously update the edge sensors based on feedback loop from the ML data greatly simplifies the way we can manage and operate a continuous self-learning and adaptive machine learning network. Which is now becoming so popular in so many areas.

You can read a more detailed description of TensorFlow Inference of Visual Images use case in Petar Torre, Solutions Architect at Intel guest blog post.

_{Figure 3 Next Generation Orchestration: Multi Domain, Sites, Clouds, Hyper-Scale}

Managing your Edge through an Internal Development Portal

Edge should be seen by the rest of the organization as a natural extension of their cloud operation. Therefore managing edge devices should follow similar best practices and tooling that are used to run any cloud infrastructure such as DevOps.

Our new upcoming release (Cloudify 7) will include a few major updates in this regard:

Simplifying the way teams can create a development and staging environment of their cloud and edge infrastructure through a self-service interface.
Continuous Updates. The new release includes a new set of built-in workflows and lifecycle interfaces that are aimed to detect continuous system failures, and configuration drifts across the entire environment while implicitly applying policies for handling those cases without breaking the continuous update process. This will simplify one of the biggest operational challenges in a distributed environment – continuous updates and day-2 operations.

_{Figure 4 Cloudify’s new consistent update across all of the automation domains}

Final Notes

It’s been roughly 4 years since we first introduced our Spire vision at MWC 2019.

Seeing the vision realized despite all the challenges and difficulties (and god knows there have been many) is quite an exciting moment for me personally and for the rest of the Cloudify team. I’m happy to say that this journey is not going to stop. Quite the contrary, it’s going to grow even faster and at a much larger scale.

Stay tuned for more announcements on this “frontier”, now that Cloudify is part of Dell.