[Subject Prev][Subject Next][Thread Prev][Thread Next][Subject Index][Thread Index]
Re: designing a brand new NFS.
Will read up on those specs...
I have no intention of reimplementing NFS exactly. The stuff I'll churn out
(not given it a name yet) will not be compatible with the original NFS in
anyway. I should have written 'a' NFS in my mail; my bad. I chose NFS (the
app this time :-) as a starting point cause it's the most popular and well
BTW, for some strange reason this thread has been hijacked by a discussion
on LILO. Please guys, next time think up a new subject line. Less confusion
----- Original Message -----
From: Arun Sharma <adsharma@xxxxxxxxxxxxxxx>
Sent: Wednesday, September 13, 2000 9:15 PM
Subject: Re: [LIP] designing a brand new NFS.
> On Wed, Sep 13, 2000 at 12:35:37PM +0530, Arsalan Zaidi wrote:
> > I've decided to take up a project implementing a simple version of NFS
> > my course at NCST. Though the requirements are simple and the time
> > I want to have a good design. Who knows, I might work on extending and
> > impementing it in full later.
> The main problem with NFS is the protocol itself. Not the implementation.
> You might consider reading about work from CMU - afs, coda etc. IBM,
> which owns the code now is releasing it open source.
> The primary problems with NFS (in no particular order):
> - No client side caching
> - Non scalability (due to 30 sec notification rule)
> - No disconnected operation
> - Lack of server replication
> Some of these are fixed in NFS v3 and v4. You might want to read up more.
> BTW, reimplementing NFS is not that hard. I did a Java implementation
> of the protocol for a class project in a matter of weeks. It was
> good enough for a demo ;)
> The mailing list archives are available at