rotaki / write_file_fast

5M write

5M を 4K ずつ書き込み逐一 fsync するのにかかった時間を計測

g++ name.cpp
for i in {1..5}; do ./a.out; done;

Intel(R) Core(TM) i9-9900 CPU @ 3.10GHz

buffer write

• buffer_write, time(ms) = 4448
• buffer_write, time(ms) = 4493
• buffer_write, time(ms) = 4436
• buffer_write, time(ms) = 4468
• buffer_write, time(ms) = 4410

overwrite

• overwrite, time(ms) = 2562
• overwrite, time(ms) = 2647
• overwrite, time(ms) = 2614
• overwrite, time(ms) = 2703
• overwrite, time(ms) = 2665

fallocate + buffer write

• fallocate_buffer_write, time(ms) = 4252
• fallocate_buffer_write, time(ms) = 4312
• fallocate_buffer_write, time(ms) = 4333
• fallocate_buffer_write, time(ms) = 4246
• fallocate_buffer_write, time(ms) = 4230

fallocate + filling zero + buffer write

• fallocate_fillingzero_buffer_write, time(ms) = 4282
• fallocate_fillingzero_buffer_write, time(ms) = 4294
• fallocate_fillingzero_buffer_write, time(ms) = 4261
• fallocate_fillingzero_buffer_write, time(ms) = 4269
• fallocate_fillingzero_buffer_write, time(ms) = 4286

fallocate + flkeepsize + buffer write

• fallocate_flkeepsize_buffer_write, time(ms) = 4321
• fallocate_flkeepsize_buffer_write, time(ms) = 4295
• fallocate_flkeepsize_buffer_write, time(ms) = 4318
• fallocate_flkeepsize_buffer_write, time(ms) = 4323
• fallocate_flkeepsize_buffer_write, time(ms) = 4343

buffer write big

5M を 5M 一気に書き込み、一度 fsync するのにかかった時間を計測

• buffer_write_big, time(ms) = 13
• buffer_write_big, time(ms) = 7
• buffer_write_big, time(ms) = 6
• buffer_write_big, time(ms) = 6
• buffer_write_big, time(ms) = 6

buffer write single fsync

5M を 4K ずつ書き込み、一度 fsync するのにかかった時間を計測

• buffer_write_single_fsync, time(ms) = 15
• buffer_write_single_fsync, time(ms) = 7
• buffer_write_single_fsync, time(ms) = 7
• buffer_write_single_fsync, time(ms) = 7
• buffer_write_single_fsync, time(ms) = 7

10G を 4K ずつ書き込み、一度 fsync するのにかかった時間を計測

• buffer_write_single_fsync, time(ms) = 10028
• buffer_write_single_fsync, time(ms) = 12022
• buffer_write_single_fsync, time(ms) = 12224
• buffer_write_single_fsync, time(ms) = 12509
• buffer_write_single_fsync, time(ms) = 12368

考察

• chunk にわけて書き込む場合 overwrite が一番早い
• fallocate を使うと通常の buffer write より少し早くなる
• fallocate の filling zero option がどれくらい効いているのかは不明
• fsync の数は少なければ少ないほどはやい
• write の数は大きな影響はなさそう