[Gluster-devel] glusterfs speed

[Anand Avati]

> I have found the same performance issues as reported in the initial
> thread.
> The the root cause of the slower performance for small files is mainly
> due to the fuse context switch overhead?
>


> If so, are there plans on the roadmap to write a proper kernel
> implementation
> of the client?
>

performance hit for small files is surely factored by fuse context switch.
But there are possible tricks around which are still possible (with a little
enhancement in fuse) like background open and dir-caching which can improve
performance considerably. There are no plans yet to port the filesystem to
the kernel, but we definitely have plans of improving small file performance
to compensate for the fuse overhead.

avati