Coda File System

Re: Mac OS X Client status

From: Greg Troxel <gdt_at_ir.bbn.com>
Date: Sat, 17 Dec 2011 09:49:21 -0500
I have long ranted that there should be a FUSE version of coda.  Not
sure if that is what "userland" means.  I think coda has overly focused
on performance, and while native-fs-speed from the cache is nice, the
first 98% of the utility function points come from it working.

Briefly, the coda kernel module basically:

  has a mini-cache of files and does reads/writes from them

  forwards operations to venus over a coda-specific interface

So instead we could have a fuse interface to venus, where all
operations, including reads and writes, are forwarded.

People in NetBSD are running glusterfs, which uses the low-level fuse
interface*, and are obtaining near-wire-speed reads from remote servers.

Later, FUSE could be extended to support a mini-cache as a generic
operation, usable by coda.  But I think on modern machines it's not that
important.

I am not clear on fuse for windows.  But at least other people are in
the same boat, and it seems like at least on the right path.

IMHO coda should target the normal FUSE interface, and be tested on
Linux, Mac (with osxfuse and fuse4x), FreeBSD, and NetBSD (with refuse,
a BSD-licensed implementation).   I think this is the single most
important thing for the coda project to remain relevant over the next
few years.

Received on 2011-12-17 10:07:16