I updated the Readme, and added a second link to the original vectorization paper. We found that most benefits came from simply switching to batch-processing; and that's why we referenced the more recent "vectorization vs LLVM" paper.
I'd say the original part of Can's work is that new ideas can be applied to PostgreSQL's robust and optimized executor, without touching any core logic. In that sense, it could also be relevant to this earlier thread on cstore: https://news.ycombinator.com/item?id=7524886
I updated the Readme, and added a second link to the original vectorization paper. We found that most benefits came from simply switching to batch-processing; and that's why we referenced the more recent "vectorization vs LLVM" paper.
I'd say the original part of Can's work is that new ideas can be applied to PostgreSQL's robust and optimized executor, without touching any core logic. In that sense, it could also be relevant to this earlier thread on cstore: https://news.ycombinator.com/item?id=7524886