RE: IEEE 802.3 Requirements

[gwinn@xxxxxxxxxx (Joe Gwinn)]

At 9:29 AM 99/5/19, Chang, Edward S wrote:
[snip]
>We do not have to worry about DMD, or very low-BW cables, which you agree.
>When users has problems with bad cables, they will find out by excessive
>errors or re-tries.  They will call field servicemen to fix for them.  Those
>servicemen will understand how to identify the bad cables to be removed from
>service.  All we have to do is to educate field servicemen how to identify
>them.

>-----Original Message-----
>From: Michael M. Salzman [mailto:msalzman@xxxxxxx]
>Sent: Wednesday, May 19, 1999 5:32 AM
>
>My experience is that on average no installer tests fiber for any thing
>other than connectivity.  To imagine that these fellows will carefully
>follow some timeconsuming and deliberate procedure to check out the cable is
>wishful thinking.  You are right that DMD is an innate 'feature' of most
>installed base fibers.  The ethernet community only recently discovered its
>ubiquity and impact.  Nevertheless we can expect the vast unwashed masses to
>blithely ignore any special handling and testing requirements untill their
>installation fails to come up and then they will search for causes.

To this (which I agree with) I would add one point:  The link startup and
autonegotiation process should be made much fussier than needed for
reliable communications over a link, so if the fiber isn't quite good
enough, the link will simply fail to work (and maybe even give a
comprehensible reason for its refusal to communicate), rather than be
flakey.

So, there would need to be at least three link failure codes, one for too
little light, another for too little bandwidth, and one for too many bit
errors.  Or the like.  One may see more than one of these kinds of failure
at once, so they should be represented as independent bits in an error
mask.  Anyway, having this kind of information will help the field
installers and maintainers to figure out which rock to look under first.

This may also be the right place to consider the issue of signal detect
(802.3z, Clause 38.2.4) , with which we had so much fun in Gigabit
Ethernet.

Joe Gwinn