Frequent ie0: No carrier messages

[John Menges]

In our experience (with Siecor's fiber-optic ethernet, especially the
passive one) this is usually caused by collisions which for some reason go
undetected or are not reported to the Sun properly.  I suspect what's
happenning is that the Sun ethernet hardware is monitoring the signal it's
transmitting (on its receive pair) to make sure it conforms to the
manchester encoding rules.  If it doesn't, and no collision is reported,
it reports that it has lost carrier.  Under older Sun operating systems,
the output errors column of "netstat -i" incremented each time a "no
carrier" message was printed.  Later versions of the operating system
didn't report "no carrier" messages each time the event occurred.  I
suspect Sun got tired of hearing complaints and changed the system so it
only reported some of the errors (e.g., only if they happen twice in a row
or some such).

There are various problems that can cause "no carrier" messages (actually,
you should monitor output errors instead, as this is now more reliable).
Our biggest contributor was the passive fiber system.  It doesn't work
because collisions are not detected reliably.  The fiber ethernet system
vendors now recognize this and have switched to active stars.  Other
contributors are faulty transceivers and transceiver cables that are too
long (50 Meters is really the limit).  You might also have a faulty
tranceiver cable, or the Sun ethernet card might be defective.  We have
eliminated nearly all output errors from our system of about 100 Suns,
except for the Sun-4s, which report output errors no matter what we do.

By the way, lots of output errors can make your nfd daemons hang.  Watch
out for this.

The problem was so big here with the passive stars that the grad students,
when voting for the year's departmental T-shirt, chose:

                         ie: no carrier

                         ie: no carrier

                         ie: no carrier

                         ie: no carrier

                         ie: no carrier

                         ie: no carrier

I have some shell scripts that collect netstat statistics from all our
suns twice a day, and generate reports sorted on various fields (e.g.,
output errors, descending).  This helped us to know where our worst
problems were, and we attacked them first.  Let me know if you want the
scripts.