Hi,

We are investigating the feasibility of providing some sort of file system
like frontend to a large data-base environment.

It seems to be very difficult to get documentation/information on how to go
about doing this.  I know about the Virtual Filesystem Interface (VFS) that
most modern kernels use, but I have no documentation on it.  Also the 
degree of vendor support for kernel access at this level is unclear.

One approach we are considering is to write our own modified NFS daemon
which would provide the NFS/RPC services over a different port than what
the regular NFS daemon uses.  Instead of reading the structure of a remote
file system, the data base would supply the information.

This method is relatively straightforward to implement and does not require
any kernel changes.  A major concern with this approach is the performance
of such a system.

Another approach would be to actually provide a simple native file system
which could provide all the services we need at a direct level.

The reason I am posting here is that AFS is a third party file system that
runs on various platforms with out the various kernels actually knowing
about it.  Information on how AFS manages to dynamically kernel load 
its own VFS extensions would be very useful.

One possibility we have considered is getting the AFS source code and 
learning/hooking into the mechanisms it uses.  The idea would be to strip it
down to just provide the type of access we need.

Is this feasible?  Are there any source examples that already exist that
do something similar?  What restrictions/copyright issues are there with
respect to AFS source?

Any references, books, experiences etc. on this would be gratefully 
appreciated!
--
Guenther Kramer                   Bell-Northern Research, Ottawa, Canada.
Internet: [EMAIL PROTECTED]
#include <std.disclaimer.h>

Reply via email to