At Brigham Young University (BYU) we have been developing a University API to expose the functionality of a reasonably generic educational institution, while consuming a very specific set of underlying technologies. Our generic institution has instructors, students, courses, classes, and locations. These resources and available HTTP methods are being combined to expose acceptable business processes such as registration, adding and dropping classes, etc. We will continue to add resources and appropriate business processes as necessary to meet our institutional needs.
Our intent is to develop future applications by consuming the University API and will encourage others to do the same. We will no longer consume the user interfaces or APIs of underlying systems. This layer of abstraction will enable us to replace the underlying technologies with new technologies that provide similar functionality. Regardless of the tools or technologies used, those consuming the University API will be unaware of the underlying change. This will give the IT organization the freedom to make changes to reduce cost, modularize monolithic applications, move to microservices, etc. without impacting application developers or end users. This will bring them freedom via abstraction.
I’m writing about this today because in my mind this is an important general architectural pattern that should be followed more often. David Wheeler, a British computer scientist, is credited with saying, “All problems in computer science can be solved by another level of indirection, except of course for the problem of too many indirections.” While most often quoted by programmers in discussions about pointers and similar constructs, I think abstraction layers, like the one discussed above, are perfect examples of additional layers of indirection that help us solve problems.
While APIs make this work easier, the approach is more generally applicable. For example, imagine you have an ERP system that is aging and the thought of living through another ERP transition scares you to death, or at least adds one more reason to consider early retirement. Imagine you add a user interface layer between the existing ERP and its users. This could require consuming an API provided by the ERP vendor, wouldn’t that be awesome, or screen scraping or via other less exciting means. When this is complete the new ERP system can be installed and connected to the user interface developed above. The two systems can be brought to a consistent state and the connected user interface can be used to keep them that way. Transaction responses can be compared until you’re confident in the new system. At this point the old ERP system can be retired. You have transitioned to a new ERP system and the users are unaware, that’s success!
There are two main points I think are worth noting. First, an additional layer of abstraction can free an IT organization to make changes without impacting end-users. Second, end-users shouldn’t use the provided user interfaces of institutionally important applications, but rather be provided with screens and applications we develop on top of APIs we control. Installation of a new application is not complete until an API we control is designed and used to create an abstracted user interface that exposes the desired functionality. When applications are installed using this model, they are more easily replaced. Freedom via abstraction!