General Electric is moving 9,000 applications to the cloud over the next three years and cutting back on the data center count from 30 to single digits. Disney has 500 projects in the cloud including customer facing applications, development, QA, websites and R&D, which is incredible progress considering Disney was just “dabbling” in the cloud one year ago. Disney has re-architected many of their legacy applications for the cloud and they are ahead of schedule.
Michael White, CTO of Disney Interactive and Consumer Products stated that they’re taking a “machine learning mindset on their road to AI.” Other large enterprises moving aggressively to the cloud include Home Depot, Colgate-Palmolive, HSBC, eBay and so on. The question is how does all this impact the CDN business model?
That’s a difficult question to answer at this point in time. However, the industry as a whole has fallen behind in machine learning (ML). The Cloud Trio (AWS + Azure + Google) has done an amazing job in building up their ML ecosystem. Machine learning might be the feature that tips the scales in the Cloud Trio’s favor where they will be able to capture the CDN spend. Some noticeable trends are the following:
- Enterprises that were shy about the cloud are now moving full steam ahead in taking their entire IT environment to the cloud
- When enterprises move everything to the cloud, that also includes CDN which by default is CloudFront and Google CDN
- The Cloud Trio continues to develop new features in volume and at a rapid pace
- The #1 goal for the Cloud Trio is to capture 100% of the IT spend including CDN services
- ML is the killer feature that can help the Cloud Trio capture all IT spend including CDN services
- CDNs and startups in the industry have fallen behind in ML based feature sets
- CDNs must make a decision fast on ML product strategy
Regardless of how challenging the competitive environment might look in the CDN vs the Cloud Trio fight, the good news – there is a global open source community dedicating themselves to developing robust ML frameworks, datasets, best practices, algorithms, and case studies for anyone to use. The one key strength that the content delivery company brings to the fight is deep domain expertise. The next step is taking the CDN domain expertise and figuring out how to leverage machine learning models to optimize everything from the software stack to personalization to predictive selling to image optimization, and so on. With the amount of raw log files being generated by even the smallest CDN, there is more than enough data to feed the data-hungry machine learning classifers so they can start predicting and self optimizing everything in its path.
Below are some of the ML features that Google and AWS have developed for their cloud customers. The best part is that these models have been fully tested, pre-trained and optimized to work out-of-box without too much configuration or coding, avoiding some of funny problems that reveal themselves during the training exercises of learning model optimization. For example, sometimes when neural networks (Sigmoid) are being trained, small changes to the weights in the input and hidden layers result in large variations in the outputs. In order to negate large variations in the models, a bias node is used to counter that problem.
Google Machine Learning Features
- Google API
- Sentiment Analysis
- Spam Detection
- Recommendation Systems
- Image Search and Classification
- Voice Search
- Churn Analysis
- Document Classification
- Purchase Prediction
Amazon Machine Learning Features
- Fraud Detection
- Demand Forecasting
- Targeted Marketing
- Click Prediction