On Fri, 3 Apr 1998 braam_at_cs.cmu.edu wrote:

> I think that weakly connected servers are presently not a good
> idea. No one has ever investigated how much bandwidth is eaten by
> resolution but it is substantial.  

As a long term goal, however, weak server connectivity would be very
useful.  Imagine, if you will, a company with a number of facilities
distributed across the world.  Rather than have the servers all in one
location and relying on each individual client to cache the right data
(and not allowing sharing of cache data between the clients), it would be
easier to have a pool of servers at each site that service the clients at
that site.  While connections between the sites would often be up, the
latency might be high, and bandwidth low.  At times, the links between the
sites would go down due to normal Internet suckiness.  While individual
clients handle this ok, it would be better if the servers handled it ok.

Example figure:

C - client
S - server cluster

(Pittsburgh, PA)
C                         C
 \                       /
  \                     /
C--S -- -- -- -- -- -- S--C
  /                     \
 /                       \
C                         C
               (London, UK)

Presumably more diverse structures would be used also.  This kind of
arrangement using a VPN is far from uncommon.

Another potential use is for a cluster of mobile machines -- in a vehicle
of some sort that has weak connectivity.  Examples might be a boat, where
satellite links exist, but are slow, high latency, and expensive.  Your 7
client machines would not have to individually manage their hoards,
rather, the server would provide a centralized hoarding service.  The same
would apply to cars, airplanes, etc.  In each case their are benfitis to a
central cache, especially if the workstations are each accessing the same
data -- also if they want to see each others modifications, which would be
useful in such an environment.  

Perhaps I am thinking of a different class of server -- rather a
centralized Venus cache and not a seperate Vice in the last described
case?  If there is clearly a "client" relationship of the cluster, that is
different from two peer server clusters where it is not clear that one is
the central copy.  So perhaps we have two pictures -- the one above with
peer server clusters, and then,

C client
S server cluster
s server


C
 \
  \
C--s -- -- -- -- -- -- S  (plus much more)
  /
 /
C


  Robert N Watson 


----
Carnegie Mellon University  http://www.cmu.edu/
Trusted Information Systems http://www.tis.com/
SafePort Network Services   http://www.safeport.com/
robert@fledge.watson.org    http://www.watson.org/~robert/