Mikhail advised me to use balanced trees instead of extensible hashing
And
I never told Mikhail that Oracle had tried implementing a filesystem
using balanced trees, and its performance was terrible leading to most
insiders in the industry concluding that balanced trees performed poorly
for filesystem File size patterns.
Interesting! From his post,
And
Of course, that filesystem exists today as btrfs.
Which, to be fair, isn’t exactly the fasted FS around. I love me some btrfs, but not for the benchmarks.