Effect of LD_PRELOAD'ing jemalloc - laurynas-biveinis/unodb GitHub Wiki

commit

unodb::db:

• micro_benchmark_n4: 7% speedup (full_n4_sequential_delete<unodb::db>/65534) to 1% slowdown (full_n4_random_deletes<unodb::db>/32768)
• micro_benchmark_n16: 5% speedup (n16_random_add<unodb::db>/4096) to 15% slowdown (full_n16_tree_full_scan<unodb::db>/512)
• micro_benchmark_n48: 1% speedup (grow_n16_to_n48_sequentially<unodb::db>/64) to 6% slowdown (shrink_n256_to_n48_sequentially<unodb::db>/512)
• micro_benchmark_n256: 17% speedup (n256_random_add<unodb::db>/64) to 0% (grow_n48_to_n256_sequentially<unodb::db>/512)

unodb::olc_db:

• micro_benchmark_n4: 2% speedup (full_n4_sequential_delete<unodb::olc_db>/65534) to 4% slowdown (full_n4_random_insert<unodb::olc_db>/100)
• micro_benchmark_n16: 4% speedup (full_n16_tree_random_delete<unodb::olc_db>/4096) to 15% slowdown (full_n16_tree_full_scan<unodb::olc_db>/512)
• micro_benchmark_n48: 3% speedup (shrink_n256_to_n48_sequentially<unodb::olc_db>/512) to 4% slowdown (shrink_n256_to_n48_randomly<unodb::olc_db>/8)
• micro_benchmark_n256: 11% speedup (minimal_n256_tree_full_scan<unodb::olc_db>/4) to 2% slowdown (grow_n48_to_n256_randomly<unodb::olc_db>/2)

unodb::olc_db concurrency: 16% speedup (parallel_get/4/2000000/process_time/real_time) to 12% slowdown (parallel_delete_disjoint_ranges/32/2000000/process_time/real_time)