>At 1:55 AM +0200 6/11/00, sthaug@nethelp.no wrote:
>>We're seeing input errors (CRC) on 3548XL switch ports connected to 7500
>>routers with PA-FE-TX. No errors are visible on the 7500 side. We have
>>tried switching cables, 3548XL switches, and PA-FE-TX on the 7500s. And
>>we still get these errors. The amount is low, about 3e-4 - but higher
>>than we would expect. Typical view:
>
><snip>
>
>I've seen a similar situation on a 7206 Fast Ethernet interface plugged
>into an HP ProCurve switch. In this case, the 7206 was logging CRC errors
>in similar proportions to what you are seeing. The HP switch was logging
>FCS and Alignment errors on RX, and no TX errors.
>
>Turns out that there was (surprise, surprise) an autonegotiation problem.
>The 7206 showed f0/0 up at 100Mbit FDX, and the HP switch seemed to think
>the interface was running at 100Mbit HDX. Setting the switch port to
>100Mbit FDX eliminated the CRC errors.
Although it may sound odd, this is most likely an example of
auto-negotiation performing as designed.
Cisco Ethernet router interfaces typically do not support auto-neg,
however most switch ports DO support auto-neg. In this case, the
auto-neg standard defines a scheme called "parallel detection" to make
it possible for an auto-neg device to deal with a link partner that
does not support auto-neg. Call the auto-neg device "A" and the
non-negotiating link partner at the other end of the link "B."
Parallel detection on "A" uses either normal link pulses (for 10 Mbps)
or the signal characteristics of 100 Mbps systems to figure out what
speed "B" is running at. However, since "B" is not sending auto-neg
signals, there is no way for "A" to determine any other
characteristics, such as the duplex setting.
According to the Ethernet standard, when "A" uses parallel detection
to determine the speed of the link partner, then "A" must assume that
the link partner is operating in half-duplex mode. In this case you
have to assume something, and half-duplex was apparently considered
the safe assumption.
Therefore, when you manually configure a non-negotiating interface to
operate at full-duplex, you need to manually configure the device at
the other end of the link to also be in full-duplex mode. Otherwise,
when left to itself the auto-neg process will result in the
half-duplex misconfiguration and link errors that you have seen.
Parallel detection only applies to 10 or 100 Mbps twisted-pair
Ethernet links. All twisted-pair Gigabit Ethernet devices are required
to support auto-negotiation, since the auto-neg system is also used to
configure some essential signaling timing characteristics over a
1000BASE-TX link.
-Charles
Charles E. Spurgeon / UT Austin Networking Services
Email: c.spurgeon@mail.utexas.edu
Ethernet Web Site: http://www.ots.utexas.edu/ethernet/
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:13 EDT