Conference netcad::hub_mgnt

>    Q1. Does anyone know of any quoted error rates for FDDI transport?
> 
>    Q2. Do the international FDDI standards include any maximum error 
>        rates that we have to meet to claim FDDI conformance?
> 
>    Q3. Can anyone offer a confident estimate of the probable error 
>        rates (they will probably be to do with quantum noise or 
>        radioactive decay in the LEDs and fibre)
> 
> The fact that they are asking these questions implies they don't believe 
> they can take advantage of any error checking built into the FDDI 
> hardware/firmware/drivers when doing multi-cast.
> 
>    Q4. Are they correct?

The 'error rate' on a 'properly configured' FDDI is very very low (far
below that of ethernet, for example), but that is not the right
question.

The question is: Does it provide reliable delivery for multicast
                 frames? 

                 I use the term 'reliable' to mean - is the sender
                 guaranteed that either the frame is delivered or an
                 error notification is returned?

The simple answer is: No.

Many normal conditions can cause one or more stations (up to and
including all of them) to not receive any frame (multicast or
otherwise); there may be buffer congestion at the receiver, the ring
may fail briefly as a node is inserted, removed, or breaks...

    Here are some answers, and some more questions. Ultimately, it is your
    customer's call as to acceptable error performance, I won't try to
    guess at what constitutes good enough. 
    
    The error rates are quoted by the FDDI PMD standard on a per link basis. 
    For minimum optical power the rate is 2.5 E-10. For nominal power 
    (nominal = >=2dB above minimum allowed receive power) the rate is
    1E-12. 
    
    For most operating rings the error rate per link usually is better than 
    1E-15. The usual experience is that the links run error free. BUT, 
    I wouldn't base a worst case design on the typical performance. 
    
    The aggregate error rate for the ring will be based on the number of 
    links.  e.g. for 100 links, with all links at 1E-12, the error 
    rate would be 1E-10. 
    
    The error checking is of two types:
    
    	Errors detected by the MAC. These include FCS errors and errors
    	caused by LOST Frames. Lost means that the frame was truncated. 
    	The MAC can differentiate an FCS error from a truncated frame. 
    
    	Errors detected by SMT. These are errors that approximate the 
     	bit error rate of the link. They are most useful for indicating
    	that there has been an error detected in the idle pattern that 
    	fills the ring in between packets. 
    
    In addition, whatever end to end error checking is in force needs to be
    considered. All the FDDI will do is indicate why there was an error -
    it won't do anything to definitively communicate that error back to the
    source. Use of the E bits of the frame won't isolate to the lost
    station. 
    
    You can read an excellent paper on this subject by Raj Jain in 
    the August 1990 issue of the IEEE Transactions on Communications. The 
    title is "Error Characteristics of FDDI". 
    
    So, you need to ask them what the allowed error rate is, the expected
    traffic, the size of the frames, the number of links in the ring, and
    the protocols that will be run end to end. If more than 1 packet is 
    exchanged per message they will also need to determine a message error 
    rate. In addition, you should consider what can happen if some FDDI
    unique events occur. For instance, there is a condition that can occur
    where the ring will be out for 15 or so seconds. It is rare, but you
    should ask what the required protocol behavior is if that type 
    of event should occur. 
    
    

Many thanks to both of you (.1 & .2) for your prompt and helpful 
replies.

At first sight they look contradictory;  .1 majors on error rates of the 
order of 1E-10 whereas .2 majors on "no reliable multi-cast".

But I think you are both saying that in the absence of hardware failures 
and disturbance the error rates are ultra-low BUT don't forget that in 
the real world hardware failures and planned re-configurations are 
fairly common.

So the OEM's position probably depends largely on his customers 
requirements during ring outages.

A fair summary?


Regards,


Martin

    
    The developers position must be that it depends on the requirements,
    and the assumption must be that FDDI does not provide reliable delivery
    of multicast (or unicast) frames, because it doesn't - it provides a
    very good best-effort.  The important question is: what is the required
    behaviour of the distributed system when frames are lost or corrupted,
    and what will it do to discover that this has occured (FDDI does not
    provide any notice that can be used for this).
    

    
    Yes, that's exactly what I hoped you would understand from my response. 
    
    Please let us know what they decide.