Skipping Load Phase - llersch/pibench GitHub Wiki

The load phase is executed single-threaded to guarantee a deterministic end result of the data structure. If the load phase takes too long, it might be helpful to preload the data structure and simply run the benchmark on a fresh working copy of the memory pool by skipping the load phase. For example, this can be achieved with something like:

# Preload the tree pool
$ ./PiBench fptree.so -n 1000 -p 0 -r 1 --pool_path=/mnt/pmem1/pool --pool_size=4294967296

# Create working copy
cp /mnt/pmem1/pool /mnt/pmem1/tmp_pool

# Skip load and run benchmark on copy
$ ./PiBench fptree.so -n 1000 -p 1000 -r 1 --skip_load=true --pool_path=/mnt/pmem1/tmp_pool --pool_size=4294967296

# Remove working copy
rm /mnt/pmem1/tmp_pool