[OpenAFS] CONFIG_ROOT_AFS or Equivilent?

[Systems Administration]

Dan Field   dof@llgc.org.uk writes:
> I'm trying to find an alternative to NFS for diskless booting linux
> workstations (LTSP). We're looking at adding a few more servers into 
> the
> mix and load ballancing them by clustering the servers with OpenMosix.
> We'd also like NFS redundancy for the read-only root filesystems and
> this is where I found pointers to AFS. I can't however find any
> reference to kernel support for root filesystems over AFS. Is it
> possible? or have I missed the point entirely here?

Since no-one appears to have answered I'll swing:

The AFS and OpenAFS architecture in particular, uses a user space 
daemon to manage much of the communications and management functions of 
AFS.  The only kernel portion is the VFS plugins needed to present the 
AFS filespace to system processes.  As such the AFS system would have 
to be redesigned to fit entirely into the Kernel in order to allow for 
an AFS_ROOT type system - and inappropriate considering the intent of 
AFS is for user space flexibility not OS specific support.

NFS is more appropriate for this function due to its long history with 
Un*x flavors and is extensively deployed in this role.  If you are 
looking for distributed access to user space files it would be best to 
use NFS for root filesystems and AFS for all userspace files, thats the 
closest you can get to an AFS_ROOT type setup.

If I understand correctly you are looking for a way to make the NFS 
root redundantly available, I have in the past implemented this with 
DNS round-robin records and soft mounting of the partitions to force 
client side timeouts.  This was with Solaris systems so YMMV - I have 
no idea if Linux would tolerate such abuse.

A better solution would be a Storage Area Network as these are 
specifically designed to solve your cluster computing problems.