Conference 7.286::fddi

The simplest answer is that we support optical bypass relays.

The more complete answer is that we also want to point out that bypass relays
are of limited use; in some network configurations you cannot use them at all,
and in every configuration there's a (fairly low) limit on how many consecutive
stations can be in bypass.  Furthermore, bypass protects against loss of power
in the concentrator, but not against disconnection of the cables, cable
breaks, etc.  In general, dual homing provides a greater level of protection
from a larger set of faults, though probably at somewhat higher cost.

There is no specific limit on the number of bypass relays.  However, there
are rules for the maximum power loss between two stations (11 dB for ANSI
multimode).  Bypass relays contribute to the loss, especially in bypass mode
(because then you are dealing with two cable runs rather than the usual one,
or more if you have several consecutive bypassed stations).  In addition, there
is also the limit on distance between stations (2 km) -- again in the case
of bypassed stations, you're adding together several cable runs.  Both these
limits, the one on loss in particular, mean that your network is unlikely to
work if more than a few consecutive stations are bypassed, and in some network
topologies it may be impossible to have ANY bypasses while conforming to the
rules!

	paul

    RE .1:
    
    In a bid specs, the customer wants to have FDDI backbone, we proposed
    DECNIS, but I am not sure whether it supports optical bypass relay.
    The spec says it is mandatory.
    
    Could you explain, Paul?
    
    Thanks and Regards,
    Fung Siong

Hi,
	The FDDI Card for DECNIS, DNSAF-XX, does support OBR.
	Note that there are SAS and DAS, MMF and SMF variants of the DNSAF.

	Mike Paquette
	Former DNSAF project engineer

    One of my favorite customers in the defense industry is planning on
    building a networked system that will be composed entirely of DAS-based
    OSF/1 machines.  They want to use optical bypass as a means to prevent
    the loss of 2 (or more) machines from segmenting the network.......
    
    While I have spent much time explaining the deficiencies of this
    approach (vs. dual homing, etc.), the customer cannot afford to
    change their configuration.  This is due to the need to meet a
    "delivered price" contractual agreement with the eventual user of
    the system.  Digital has been called in to replace the "as bid"
    solution that used IBM gear, token ring, etc....hence their price
    constraint.
    
    Anyway: Could someone help me to better understand exactly when
    a bypass would switch out the local FDDI station?  I can see where
    the loss of power to that station would do it....but are there any
    other conditions?  (like cpu halt, FDDI interface "broken", etc.)
    
    Conversely, are there any failure conditions of the controller that
    would not cause the bypass to switch...and therefore break the ring?
    
    One of my (possibly naive) assumptions is that as long as the FDDI
    interface has power, it will correctly participate in the FDDI ring.
    Nothing other than the microcode in the controller is necessary.....
    I don't care that the controller is not actually sending/receiving
    any packets of it's own..but that it is acting as a "repeater" of
    other FDDI packets.
    
    -tnx,
    	Jeff

Two points...

First of all, the dual ring recovers (by "wrap") from any SINGLE fault.  The
benefit you get from bypass relays is that SOME multiple faults are also 
covered.

The bypass relay is controlled by the adapter firmware.  If either port is
active, the bypass relay is inserted.  If the adapter is turned off, the
bypass relay will be switched to bypass.

When the adapter is active, it repeats any frame it did not send.  So even
if the host is halted, that won't disrupt other nodes on the ring.

Most adapter malfunctions would be caught by other mechanisms, such as failure
to complete the connection startup (PCM) or by PC Trace.  The bypass relay
is not involved in these.  If you have a problem of that sort, you will
probably end up with a wrapped ring.  If two adapters have such a problem,
your network is partitioned.

As you said, dual homing is a better choice, for several reasons: (1) it avoids
the link power problems you get with multiple stations in bypass; (2) it cleanly
handles malfunction cases like the ones I mentioned.  For defense systems
these concerns are particularly critical, which is why Safenet-2 moved away
from bypass relay based designs to a dual homed (or multiple rail) approach.
However, with limitations, the bypass relays can provide a useful function.
They certainly are likely to be better than a dual ring without bypass relays.
It's too bad, though, that the customer is in this contract bind and ends up
having to provide second best to their customer.

Incidentally, there's one unique drawback of bypass relays for this market:
being mechanical devices, they are vulnerable to shock; some shocks may 
disrupt the ring momentarily (causing lost packets) where a configuration
without them would have seen no effect.

	paul

    
    ...as always, many thanks, Paul!
    
    We are continuing to try and convince the customer to *not* use
    optical bypass.  I am hoping that we can force a reading of the
    functional specification for failure modes that will allow us to
    bring a sane configuration into this.
    
    -tnx,
    	Jeff