From Software-Defined to Metadata-Driven

Software is eating the world! Every company is becoming a software company. If companies don’t, they cease to exist. Just imagine: you are a thermostat maker and suddenly you have Google as a competitor (via its Nest acquisition). This is just one of the many recent examples.

Interestingly a lot of the innovations in the software industry are fuelled by abstraction and automation, concepts that are well-known in the Model-Driven Development (MDD) community. As the world is awakening to these concepts there is a clear opportunity (and need!) to bring MDD to a much broader audience.

Moving towards a metadata-driven software stack

In my keynote at Code Generation 2014, titled “Software is eating the world and MDD should be in the driving seat”, I analysed what’s happening on all layers of the software stack. We explored how we can become more relevant as an MDD community. There is a ton of knowledge and experience in our community that could move the needle for a lot of companies, but are we using it?

There are three ways in which Model-Driven Development can make a difference. The gist of all three of them is that we need to move from “software-defined” to “metadata-driven”. Let’s have a look at them:

  1. Model-Driven PaaS: In todays cloud architectures we are applying abstraction and automation on each layer of the stack (for more background on the layers I use in the presentation see my earlier article that introduced this framework for describing the cloud landscape). However, the average Platform-as-a-Service (PaaS) vendor has a sole focus on abstraction and automation regarding deployment. Most PaaS vendors do not have a vision on application development, they use the same languages as we are using for years, while most of the time spend in the application lifecycle is on developing and changing the code. That’s why I propose to add a layer on top of PaaS that is focused on Model-Driven Development: Model-Driven PaaS (sometimes dubbed “high-productivity PaaS”). In a Model-Driven PaaS the applications are specified using models or metadata, developing applications is metadata-driven.
  2. Metadata-driven deployment and operations: if we have a Model-Driven PaaS layer we can also do more from a deployment and infrastructure perspective. We could derive all deployment meta information and non-functional characteristics from the model and automatically generate the right configuration for lower layers. Again: metadata-driven! Examples are:
    • Automatically adjusting the infrastructure when the Service Level Agreement (SLA) is not met (e.g. elasticity of resources, deploying close to users with high latency).
    • Automatically configuring the network based on integration points specified in the model/metadata of the application.
    • Characteristics and use of entities in the app model should lead to auto provisioning of the right database (type, size, scale).
  3. APIs and data as first-class language concepts: if we annotate APIs and data with proper metadata they can become an integral part of the app model. APIs and data extend the (modeling) language with “build-in” knowledge. The Wolfram language is an example of this concept. Another example is the concept of Mendix App Services: if an app service is included in the application model the data and logic from this app service can be used as if they are part of the application model itself.

The slides of my presentation

Below is the complete slide deck of my presentation. Please share your thoughts on this subject!