I find the name data-oriented design very confusing – it sounds like data-driven development (letting hard data determine decisions in development) or data-driven programming (control flow determined by loaded data and not hard coded), except it’s something completely different: designing for current cache-dominated CPU architectures combined with an emphasis on making a few highly-reusable functions.
It seems to me that there are a number of other names that better convey what it actually is:
- (Cache|[CPU] Architecture|Layout)-(Oriented|Optimized) (Design|Development|Programming): that’s 24 options that all convey that it is about optimizing your coding for the way CPU caches work.
- Database-inspired design: This implicitly conveys the cache-optimized design, as databases are often optimized for that, and also suggests the use of the reusable transforms on the data. It seems to me, based on the data-oriented design book, that this is what the approach actually is as well, though I am more partial to the above as they convey why it actually works (because the way your data is arranged in memory lends itself to efficient prefetching of data and tries to minimize the number of hard to predict branches in the instructions that process it).
5
For a moment I was going to ask what the question actually was, then I remembered it was in title.
I agree “data-oriented design” is a pretty obtuse and uppity term.
The term seems to originate from being defined in contrast to “object-oriented design” where an object is thought of as code and data defined together in source code and handled together at runtime.
Proponents of DOD basically argue that OOD is inefficient in the execution (their area seems to be game development and high-performance C++), and (underneath a veneer of new-fangled terminology) advocate a return to old-style procedural design concepts where data structures exist separately from the algorithms which operate on them, though the two are still designed to mate together efficiently, and where programs are designed in ways that are most efficient for execution on a particular computer architecture (rather than conforming to generic OOD diktats).
In briefly researching this, I also noticed terminology used like “array of structures” and “structure of arrays”, which basically correspond to what is known as “rowstore” and “columnstore” in database engine contexts (also frequently “row-major” and “column-major” more broadly), and which is all concerned with the same issues about how maximise the locality of storage for certain access patterns.
In essence, DOD is just not-OOD.