"M. Satyanarayanan" <satya_at_cs.cmu.edu> writes:

> In FUSE, is there a way to direct read() and write() operations
> directly to a local file, without bouncing up to user space?  That's
> what the Coda kernel module does today, and we need to think through
> the performance implications if that is not possible.  The latency and
> bandwidth of read() and write() operations are not slowed down by
> Venus today --- only open() and close().  Will we be giving that up?
> If we have to, will it matter?

I'm not really sure.  It certainly seems possible to add in support for
that kind of redirection to FUSE (or to puffs, which I think is more
capable than FUSE, being a second system :-).  With page loaning and
read ahead, I'm not sure how much this will actually hurt.  It would be
interesting to get coda/FUSE running and benchmark it.

But, from where I sit, this is a third-order performance question, and
"coda doesn't work on my mac" is 10^6 times more important.  Were there
working coda kernel support for all the places FUSE works, then we
probably wouldn't be having this conversation in the same way :-)

I don't think people are advocating "rip out coda kernel module support
From venus".  It's more "add in a way to compile for FUSE instead".