Since FUSE 3.17, there's a new option called FUSE_PASSTHROUGH that allows letting the kernel handle entirely in kernelspace file read and writes, significantly reducing the overhead of FUSE filesystems and increasing the performance.
Tests from the patch author show a very, very significant increase in speed:
Results are presented in the following table:
+-----------+------------+-------------+-------------+
| MiB/s | fuse | passthrough | native |
+-----------+------------+-------------+-------------+
| read | 471(±1.3%) | 1791(±1.0%) | 1839(±1.8%) |
| write | 95(±.6%) | 1068(±.9%) | 1322(±.8%) |
| randread | 25(±1.7%) | 860(±.8%) | 1135(±.5%) |
| randwrite | 76(±3.0%) | 813(±1.0%) | 1005(±.7%) |
+-----------+------------+-------------+-------------+
(There's also upcoming work on in-kernel passthrough for other operations but those still haven't landed AFAIK)
I think bindfs would benefit a lot from implementing these changes. I think mechanism would interfere with rate limiting, but it would be possible I guess to use it if rate limiting is not enabled.
Since FUSE 3.17, there's a new option called
FUSE_PASSTHROUGHthat allows letting the kernel handle entirely in kernelspace file read and writes, significantly reducing the overhead of FUSE filesystems and increasing the performance.Tests from the patch author show a very, very significant increase in speed:
(There's also upcoming work on in-kernel passthrough for other operations but those still haven't landed AFAIK)
I think bindfs would benefit a lot from implementing these changes. I think mechanism would interfere with rate limiting, but it would be possible I guess to use it if rate limiting is not enabled.