bulk_fiba_artifact_guide.md

Bulk FiBA Artifact Guide

There are a few relevant folders:

• cpp/src/ - all C++ source files, importantly FiBA.hpp and AMTA.hpp.
• cpp/ - the Makefile to build the benchmark drivers.
• experiments/ - helper scripts to run and process data from experiments.

Compilation Instructions

The code can be compiled using just make, a recent g++, and a recent libboost installation. Our experiments took advantage of mimalloc; this is not necessary but can deliver lower latency variability in memory management calls.

To compile the code, change into cpp/ and run make like so:

make bulk_evict_benchmark bulk_evict_insert_benchmark bulk_data_benchmark

This will build the relevant binaries and store them in cpp/bin/.

Optional: If mimalloc is desired, follow the instructions in README.md therein.

Running the experiments and parsing the results

The scripts in experiments/ that pertain to bulk operations match the pattern run_bulk*.py, all written in Python 3. An example of how all these are called can be found in run_all_bulk.sh in the same folder. These scripts store experimental results in experiments/results.

For real-world data experiments, datasets are expected in the folder experiments/data/. The accompanying paper uses the NYC Citi Bike dataset. Once the data is downloaded and concatenated according to the instructions in the paper, it is stored as experiments/data/NYC-Citi-Bike/catted-citibike-tripdata.csv.

The processing scripts to parse the results are process_bulk*.py. They read from experiments/results. These scripts expect pandas, matplotlib and numpy, and generate figures in figures/.