LMDB benchmark testing sorted vs unsorted insert order for a hashtable

lmdb-bench.cpp

#include "precompiled.hpp"
#include "stopwatch.hpp"
#include "lmdb++.h"

int main() { //genbuild {'entrypoint': True, 'ldflags': '-llmdb'}
	lmdb::env env = lmdb::env::create();
	env.set_mapsize(1UL * 1024 * 1024 * 1024 * 1024);
	env.set_max_dbs(64);
	env.open(std::string("/home/jbosboom/scratch/benchmark.mdb").c_str(), MDB_NORDAHEAD); //TODO: flags?

	lmdb::dbi hashtable;
	{
		lmdb::txn txn = lmdb::txn::begin(env);
		hashtable = lmdb::dbi::open(txn, "gadget_hashtable", MDB_CREATE | MDB_INTEGERKEY);
		txn.commit();
	}

	std::mt19937 rng(1235);
	std::uniform_int_distribution<unsigned int> length(15, 300);
	std::uniform_int_distribution<std::size_t> hash(0);

	const bool sort = true;
	const std::size_t batch_size = 5000;
	std::vector<std::byte> data; //we're not bothering to mutate the data here (shouldn't matter)
	data.resize(1024);
	std::vector<std::size_t> hashes;
	hashes.reserve(batch_size);

	Stopwatch stopwatch = Stopwatch::process();
	for (unsigned int batch = 0; batch < 100; ++batch) {
		for (std::size_t i = 0; i < batch_size; ++i)
			hashes.push_back(hash(rng));
		if (sort)
			std::sort(hashes.begin(), hashes.end());

		auto txn = lmdb::txn::begin(env);
		{
			lmdb::cursor cur = lmdb::cursor::open(txn, hashtable);
			for (std::size_t hash : hashes) {
				std::string_view value(reinterpret_cast<const char*>(data.data()), length(rng));
				cur.put(lmdb::to_sv(hash), value, MDB_NOOVERWRITE);
			}
		}
		txn.commit();
	}
	Stopwatch::Result elapsed = stopwatch.elapsed();
	fmt::print("{} {} {}\n", sort, elapsed.hms(), elapsed.userMillis(), elapsed.systemMillis());
}