Search on Sorted Data

Search on Sorted Data (SOSD) is a new benchmark that allows researchers to compare their new (learned) index structures on both synthetic and real-world datasets. It is provided as C++ open source code that incurs little overhead (8 instructions and 1 cache miss per lookup), comes with diverse synthetic and real-world datasets, and provides efficient baseline implementations. Here we provide an interface and leaderboard for index structures on our own hosted SOSD benchmark.

You can clone the original repository here to run benchmarks on your own machine. This dataset is distributed under the GPL 3.0 License.

To have a new learned index added to the benchmark and leaderboard, send an email to Andreas Kipf or Ryan Marcus with the manuscript and repository of the index.