It's day two at the Analytics and Data Summit at Oracle's HQ. Dr. Frank Munz gave an excellent presentation on "Serverless Computing and Machine Learning."

IMG_0576.jpg

He went over the pillars of what Cloud needs:

  1. API
  2. Elasticity
  3. Pay Per Use

 

He went on to cover that all the above give you a fully programmable data center. However, when you have a situation where eg your IT Data Center or your Compute Instance in the cloud usage is cyclical, Serverless solutions are a better remedy. Case in point is Netflix. They have a lot of demand on their streaming services in the evening but not so much in the mornings.  Dr. Munz mentioned that Serverless came first to play with AWS Lambda in 2014. Lambda is a Function as a Service which allows a user to trigger functions based on events without thinking about servers, containers or language runtimes. It takes Cloud to the next level and truly automates the elasticity and true pay per invocation.

 

Although Lambda was a great start in 2014, it suffers from some inadequacies, in that it comes with a vendor lock-in because it is not standards based. There are a couple of Serverless frameworks including the Fn Project which is Open Source and Polyglot, so can use any language, Java, Python etc.

 

To give a demonstration of Serverless application, Dr. Munz bridged to Machine Learning. An example being calculating Airline delays. At present Fn is not available as a service in the cloud but you can still run it as an instance on a Server. There are potentially one to two orders  savings in cost using it. If you provide Fn your Docker Login, it can push the Docker Container to the Docker Hub. Then it can pull the Container from the Registry and run it. Examples of apps could a Recommendation Engine, or the specific demonstration he gave was colorizing a photo. Dr. Munz took a photo of the session attendees, grayscaled it  and then ran it on Open FaaS. The results were amazing and rather than go on about it, see for your self.

IMG_0585.jpg

 

Dr. Munz concluded by asking attendees to consider "hosting your prediction model on FaaS."  The Fn project gives you both Function and Containers and that a future Fn Cloud Service would provide:

  1. True Pay per Use
  2. Automated scaling
  3. Integration with other Oracle Cloud Services
  4. Standards based.

That's it!