I've been a big believer in AFS from the 80's. But there are two big
obstacles that I have seen historically that have gotten in the way of
AFS achieving ubiquity. These are my opinions, and may not reflect
reality.
1. AFS is more complex to install and administer than NFS. Most sites
get their OS with NFS pre-installed, and don't have to think very hard
even to bring a server online. Microsoft will package their network
filesystem in the same way.
2. Transarc has established a pricing, and product placement policy
that focuses on a narrow range of customers where Transarc can make the
maximum number of high dollar sales per unit support. Transarc does not
have the support infrastructure to answer calls from all and sundry
people on all and sundry platforms. Consequently they cannot get into
the business of supplying THE filesystem of the Internet.
I think obstacle 1 could be removed if three things were done:
1. Rethink the server model and do the technological changes to make an
AFS cell trivial to bring online. To bring up an NFS server, you set a
couple bits, and your existing files are exported. What AFS requires in
comparison is **HORRENDOUS**. Current thinking is that all that
complexity is necessary for efficient operation and for providing all
the functionality. Someone with absolutely no taste needs to kludge
together an efficiency/functionality compromized dumb server setup mode.
2. Package the client and the server systems more simply so that it's as
easy to layer in AFS as it is to layer in NFS on an OS distribution.
3. Transarc should negotiate strategic alliances with OS vendors to
offer AFS clients as a low-cost as a pre-installed option like was done
at about this same phase in NFS's history.
----
To remove obscacle #2 Transarc should establish a strategic relationship
with a 3rd party end-user service provider, and have that 3rd party
triage calls from AFS client end-users. Then the client licensing price
should drop to as near zero as possible, to motivate a big customer
base. Then AFS servers should be positioned as a premium product that
all right-thinking data centers would have.
This model seems to me to be very much like the one that seems to have
gotten Netscape installed and supported on a delightfully ubiquitous
basis. Naturally Netscape client/server and AFS client/server are
different situations. I think though the differences favor AFS: AFS
functionality has been around and debugged more than browser
functionality. The server side is a lot more difficult to code and get
right with the AFS protocol than the http protocol. The complexity and
value added occur naturally at the server end leaving the client end
simple and not in need of continual refresh by giving away many versions
of progressively complex software! The key? GIVE THE CLIENT AWAY FOR
FREE.
I'd like to be the first to shake the hands of the Transarc principals,
if this work is already in progress, and will congratulate them on
having taken over when this strategy.
-Bill Cattey
MIT
AFS user since 1988
