Course project for FML. Code mainly adapted from https://github.com/learnedsystems/SOSD.
Two options:
- To build from scratch:
bash ./scripts/generate.sh # generate data
# IMPORTANT: clone the https://github.com/learnedsystems/RMI repo and replace the path below
bash ./scripts/build_baseline_rmi.sh /path/to/the/RMI/repo- To download prebuild data and baseline:
bash ./scripts/download_prebuild_data.shTo benchmark all the baselines:
bash ./scripts/benchmark_baseline.shEach baseline will be run 5 times on each query workload. Result will be saved to run/result/ with each line of format as follows:
<model_name>,<variant_id>,<run_time_1>,<run_time_2>,<run_time_3>,<run_time_4>,<run_time_5>,<model_size>,<build_time>,<last_mile_searcher>,<workload>
For example:
RMI,3,1343.46,927.907,877.172,919.756,937.959,24080,0,BinarySearch,wiki_ts_200M_uint64_queries_10M_in_test
RMI,4,1190.39,649.651,669.04,648.392,668.077,240080,0,BinarySearch,wiki_ts_200M_uint64_queries_10M_in_test
RMI,5,1030.91,425.019,429.79,398.885,398.248,2400080,0,BinarySearch,wiki_ts_200M_uint64_queries_10M_in_test
RMI,6,559.205,371.127,352.677,352.18,344.074,24000080,0,BinarySearch,wiki_ts_200M_uint64_queries_10M_in_test
To test your own RMI model, please refer to src/benchmark.cc. For example:
./build/benchmark ./data/wiki_ts_200M_uint64 ./data/wiki_ts_200M_uint64_queries_10M_in_test 5 rmi /path/to/your/rmi/parameter/dataThis will use query workload ./data/wiki_ts_200M_uint64_queries_10M_in_test to benchmark the RMI model with parameters /path/to/your/rmi/parameter/data.