Random sampling from B + trees
Frank Olken, Doron Rotem · Very Large Data Bases · 1989
We consider the design and analysis of algorithms to retrieve simple random samples from databases. Specifically, we examine simple random sampling from B+ tree files. Existing methods of sampling from B+ trees, require the use of auxiliary rank information in the nodes of the tree. Such modified B+ tree files are called “ranked B+ trees”. We compare sampling from ranked Bt tree files, with new acceptance/rejection (A/R) sampling methods which sample directly from standard B+ trees. Our new A/R sampling algorithm can easily be retrofit to existing DBMSs, and does not require the overhead of maintaining rank information. We consider both iterative and batch sampling methods.