The rapid pace of technology innovation is simply proving too much for some vendors. With so many new technologies hitting the market at once, as in those technologies that become the building blocks for new features, trying to build the right product feels like a life or death sentence. Build the right product and fortunes are made, pick the wrong one, and it’s the opposite.
Cloud security breathed new life to the CDN industry, enabling some to build profitable, value-added services. Edge computing is having the same impact, its helping vendors transform their business models, and some are adding new, AWS-like features that sell for a premium. Although we’re in the beginning stages of the edge computing lifecycle, Kubernetes and container-based technology have entered the picture. Some vendors have embraced it, while others are refusing to do so because it doesn’t fit the mold, in that containers will require a different set of servers. And if that wasn’t enough, now we have artificial intelligence.
Cloud security, edge computing, and Kubernetes are transformational technologies, in that they can change the direction of an entire industry, as it’s doing to the CDN industry. AI is at an entirely different level, nothing comes close to the potential impact it will have in the world.
AI is no longer an abstract technology that requires a Ph.D. in mathematics to work with. In the last two years alone, AI has gone from very complex technology to an easy-to-use system where programs can be written in Python and sophisticated neural networks can be spun up in the cloud. Today, all it takes is the right set of tools, a few lines of code, and lots of patience, where any savvy engineer will be able to set up a powerful deep learning neural network. At the pace that AI is evolving, it won’t be a surprise if there’s an AI program in two years that does all that for you – select the right tools, write the algorithms, and so on.
A global community has evolved around AI and they are simplifying the process of working on this complex technology. In fact, there is no longer a need to write algorithms from scratch, because many have been created by the community. And no company has contributed more to simplifying AI than Google. Google has introduced a slate of products in this area, the most notable being TensorFlow, a powerful, open source machine learning framework that comes with a comprehensive set of tools, libraries, and best practices to build ML models. Who uses TensorFlow?
- Coca-Cola, Dropbox, eBay, Airbus, Uber, Qualcomm, GE, Intel, Twitter, LinkedIn, AMD, and many more
It’s not only the large enterprises using AI, but startups in the areas of cloud security, network analytics, computing, and more. AI-driven startups are in the hundreds.
AI is an umbrella term, whereas machine learning (ML) is a subset of AI. ML is basically a system that is able to learn from large volumes of data, identify patterns via algorithms, then make predictions. If predictions are inaccurate, then it tries again until it gets it right. There are three broad types of ML:
- Algorithms: regression and backpropagation
- Neural networks: deep learning
- Algorithms: K-Means and K-Medians
We just mentioned a few algorithms above, but in reality, there are literally dozens of different types of algorithms, if not hundreds. Today, deep learning has taken off and TensorFlow is one of the reasons for it. Josh Gordon, a Tensorflow (TF) evangelist made the following statements:
- TF powers every ML project at Google
- TF has large API stack
- All the code runs on Colab, a free-based Jupiter notebook that comes with free GPU
- Been around for 3 years and it has 2,152 contributors and growing
- Powered by C++ backend and almost everyone writes Tensorflow programs in Python
- Four API styles: Keras, Esitmators, Eager execution, and Deferred execution
- Keras is an API that is used for training and defining neural networks using lego-like building blocks
- Keras is an extremely popular, user-friendly machine learning library written in Python that works on top of TF
Google’s core mission for machine learning is to simply the heck out of it so anyone and everyone can build sophisticated training models and ML programs.
CDN and Infrastructure Industry
Although machine learning tools like TF are no more than 3 years old, it’s transforming various technology industries. Take Anodot for example, this startup has developed a next-generation APM / analytics service that uses supervised, cluster-based algorithms to deliver infrastructure insight to technical teams and business insight to non-technical execs in areas such as products, customer service, cost, revenues, etc.
In short, AI is upending the analytics industry. Another industry being impacted by AI is security. Nowadays, if a security startup emerges from stealth mode without AI at its core, that’s unusual.
However, one sector that hasn’t been impacted in the slightest way is the CDN industry. The day is coming when AI will disrupt the CDN business model. Any vendor that fails to invest in AI will fall behind.
Incorporating AI into the CDN tech stack is easier than it was two years ago, as there are a plethora of ML tools that can be used as building blocks for creating features, and these tools include scikit-learn, theano, caffe, Torch, Accord Framework, and CNTK, just to name a few.
The question for CDNs now – how can AI be used to build premium products that command premium prices. If we break down the tech stack into parts, we can identify those areas ripe for disruption.
CDN Tech Stack
- Content delivery
- caching, storage, DNS, mobile app delivery,
- Global network
- Video streaming
- DDoS protection
- Bot mitigation
- API security
- Edge Computing
- Functions / Serverless Scripting
- CDN Database
- Kubernetes / Containers / FaaS
Based on the list above, security would be a great place to start. Incorporating machine learning such as clustering algorithms can help vendors identify legitimate users from illegitimate. And as time goes on, the ML program will continue to learn and improve on its own, as it is fed with a continuous, large volume of data.
Another area that can benefit from ML is global network performance. Improving performance by squeezing milliseconds out of the network is a long life quest for all CDNs. What if you had an intelligent network that could predict congested paths before they become congested, and route traffic accordingly. The list goes on, but that’s a topic for another time.
Is the AI-driven CDN coming soon? Nah, everyone is too wrapped in edge computing at the moment.