I am mostly interested in two things for reading. Therefore, the language binding provided fake file systems provide following:
-
file reading performance
- 'infinite' file; can be used to play with dd
-
directory iteration performance (TBD)
| backend | read perf |
|---|---|
| native ( /dev/zero ) | 16GB/s |
| go-fuse | 315MB/s |
| python-llfuse | 165MB/s |
Also, testing with real 'testfile' (raspbian installation image) on my laptop;
| backend | read perf |
|---|---|
| native | ~900MB/s |
| go-fuse | 290MB/s |
| python-llfuse | 115MB/s |
It seems that Python binding uses at most 4kb block size, even with appropriate mount options, on OS X. It renders it relatively .. slow.