Conference 7.286::fddi

        DECconcentrator
          |         |
      DECbridge     ------------------------------ DECbridge 618
        510               single mode fiber
                           8 micron - 28km

Both are legal.  There are other possibilities too.  Which is best depends
on circumstances.  Did you show the entire proposed FDDI, or just a piece?

a. If you connect a number of things to one concentrator, all those things
   are dependent on the health of that concentrator.
b. If you don't like that, you can use DAS things and "dual homing" to two
   concentrators.
c. Alternatively, a DAS can be directly connected to the dual ring.  The dual
   ring copes with single faults (though not double faults).
d. To minimize the probability of global disruption due to a double fault, 
   the general rule is to keep the dual ring small, both in component count
   and physical size.
e. Don't apply rule (d) blindly.  Mentally insert some probable faults into
   various configurations; you may find that a particular configuration is
   just as reliable with a long dual ring as with the alternatives.
   (I've helped design some configurations where a dual ring, with single mode
   fiber, ran between two towns.  That was a special case, but it serves to
   illustrate the fact that you shouldn't blindly follow a "stay away from
   dual rings" attitude...)

By the way, if what you showed really WAS meant to be the whole network,
then I'd say the variant in .0 is best (more reliable and cheaper).  There's
another variant that's cheaper yet, but VERY non-expandable: connect the
DECbridge 518 directly (no concentrator) to the 618.  That's the special case
two-node FDDI, built from two SAS directly connected.  To expand it beyond
two nodes you need a concentrator.  (Compare that with the two-DAS case,
which expands to more than two nodes just by adding the additional nodes,
subject to the caution under (d) above.)  The two-SAS case is more vulnerable
to cable problems than the two-DAS configuration.  If all your fibers are
in the same cable, the same backhoe will cut all of them, so that may not
be a consideration.  (Then again, if the traffic is important enough to
send over single-mode spans like that, you probably do care about backhoes
and should consider having the redundant fibers be physically separate!)

	paul

Paul,

Thanks for the ramblings, what I gave is the whole network as it will exist for
today. I was confused about the ability to connect two SAS together. In the 
network buyers guide it states that a SAS could only connect to a concentrator 
or did I miss a exeception somewhere. The customer would really rather do
a two SAS connection.

Thanks again,
Brad

A two SAS FDDI is somewhat like a two-endnode network.  You can build it,
but to expand it you need an extra box (concentrator in the first case,
router in the second).  So you have to be careful when discussing that case
to make sure the customer understands the limitations.  But when you have
a special case that doesn't call for any expansion, or an experimental
setup where the cost impact of expansion is well understood, it makes for
an attractive, lower-cost solution.

	paul

   Paul Konig has several other times mentioned that a minimum ring can be
   build of two SAS stations, but expansion beyond that would require a
   concentrator.  I'm wondering just what breaks (or if it does break) when a 
   3rd SAS is patched in.  I understand that this can't be done with standard
   cables, but if I had the proper bundle of single fiber jumpers with the
   proper connectors, I could certainly connect up 3x SAS stations or more in
   a ring and have at least the physical media connected among them in a legal
   (?) fashion.  In the spirit of 'yeah, we know we'd never want to operate
   this way, but for staging/testing...', would this work?  If yes, wouldn't
   it be transparent to the devices, drivers & applications?  If no, what am I
   missing ?
    
   Jim Timmerman

    This sort of configuration is discussed from time to time; it even has
    a name ("Class C Ring" -- after "class A" which is the dual ring, and
    "Class B" which is the concentrator tree).  
    
    But it is NOT part of the standard.  The standard very clearly requires
    that station connections are duplex, i.e., if the transmitter on a port
    connects to remote station port X, then the receiver of that port must
    also be connected to remote port X.  This requirement is fundamental to
    the design of a number of the FDDI protocols.  For example, PCM
    requires it (and indeed has some steps that detect SOME cases where
    this requirement is violated).
    
    However, if you were to plug such a configuration together, it is
    possible that it will "work".  This will depend on timing; the
    resulting setup will certainly not be reliable.  It would require a
    DRASTIC change in PCM -- for starters -- to do more than that.
    
    So the short answer is "NO WAY".
    
    	paul