echo <number-of-1GB-hugepages> | sudo tee /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages
- gcc
- cmake
- python
- cpplint
- glog
- gtest
- tbb
- jemalloc
mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release # RelWithDebInfo, Debug
make -j`nproc`Hand-crafted:
btree_olchashtable_olc
Tabular:
btree_inline_tabularhashtable_inline_tabular
Naive-OCC:
btree_inline_tabular_materializedhashtable_inline_tabular_materialized
Naive-MVCC:
btree_tabularhashtable_mvcc_tabular
Pessimistic Locking:
btree_lc_stdrwbtree_lc_tbbrwhashtable_lc_stdrwhashtable_lc_tbbrw
C: Lookup-onlyG: Lookup-mostlyA: BalancedH: Update-mostlyF: Update-only
# Volatile
$BUILD_DIR/benchmarks/ycsb/ycsb_<variant>_bench_256 --workload <workload> --threads <number of threads> --records <number of keys> --duration <seconds>
# Persistent
$BUILD_DIR/benchmarks/ycsb/ycsb_<variant>_bench_256 --workload <workload> --threads <number of threads> --records <number of keys> --duration <seconds> --persistent --log-dir=<logging directory>
# Persistent (No I/O)
$BUILD_DIR/benchmarks/ycsb/ycsb_<variant>_log_nop_bench_256 --workload <workload> --threads <number of threads> --records <number of keys> --duration <seconds> --persistent --log-dir=<logging directory>$BUILD_DIR/benchmarks/tpcc/tpcc_masstree_bench --duration <seconds> --threads <number of threads> --scale-factor <scale factor>
$BUILD_DIR/benchmarks/tpcc/tpcc_tabular_btree_bench --duration <seconds> --threads <number of threads> --scale-factor <scale factor>To minimize NUMA effect, run benchmarks with numactl --membind 0 or numactl --interleave 0,1,..,<N> when threads are within one socket or across N sockets.
Every benchmark binary accepts --help to print complete available options.