Monday, October 17, 2022
Saturday, October 15, 2022
One of the things I've been wondering for a while now is how the performance of macOS's EXFAT driver is representative of its peers. It's notably slower than what you would see in NT, but not so bad until you go from the sequential 1M to random 4K part of my choice benchmarks. Once you hit the randoms, it goes form "I wonder if that's lack of optimization in the driver, or the I/O system design" to abysmal. But to be fair that is the worst performing metric anywhere, and I'm more interested in the sequential performance.
Well, having a nice shiny (or should I say, mat?) Samsung T7 Shield that was on sale, I decided to do a little test cycle. EXFAT, FAT32, HFS+, and APFS. This drive is designated for Time Machine duty, so I have no need for it to remain on a interoperable file system.
Using AmorphousDiskMark 4.0. EXFAT as formatted out of the box: Test - Read MB/s Write MB/s SEQ1MQD8 - 586.42 691.32 SEQ1MQD1 - 594.45 690.05 RND4KQD64 - 21.75 13.68 RND4KQD1 - 21.70 13.48 FAT32 as formatted MS-DOS (FAT32) from Disk Utility: Test - Read MB/s Write MB/s SEQ1MQD8 - 516.03 690.32 SEQ1MQD1 - 596.97 691.80 RND4KQD64 - 21.56 13.64 RND4KQD1 - 21.50 13.51 HFS+ as formatted Mac OS Extended (Case-sensitive, Journaled) from Disk Utility Test - Read MB/s Write MB/s SEQ1MQD8 - 612.39 820.77 SEQ1MQD1 - 578.25 691.00 RND4KQD64 - 120.48 55.44 RND4KQD1 - 18.33 14.70 APFS as formatted APFS (Case-sensitive), after converting from MBR to GPT from Disk Utility. Test - Read MB/s Write MB/s SEQ1MQD8 - 733.22 818.84 SEQ1MQD1 - 617.40 684.06 RND4KQD64 - 121.67 55.13 RND4KQD1 - 21.27 13.83
This makes me suspect the performance lossage is more to do with how optimized the FAT drivers are. I should really repeat this with one of my USB flash drives where the performance sucks to begin with, but I don't want to spend all day on this :^o).