Troubleshooting Guide

concern about the link going up/down on NICs with the Windows 2000 or Windows ME operating
systems, disable the power management feature as a first step in order to troubleshoot link
up/down situations.
 Disable the NIC power management functionality. Many NICs support their own power
management capability.
When you troubleshoot link up/down issues, disable this feature. For information on how to disable
power management, refer to the NIC documentation.
 Adjust switch jitter tolerance.
Jitter tolerance, based on the IEEE 802.33u-1995, clause 25, must not exceed 1.4 nanoseconds.
However, there are situations in which NICs that operat out-of-specification with respect to
excessive jitter cause link up/down situations on Catalyst 6000 and 6500 10/100 ports. The
workaround for this issue is to increase the jitter tolerance on the Catalyst 6000 and 6500 switches
for 10/100 ports to 3.1 seconds. The set
port debounce mod/port enable command enables the
feature. As an ultimate solution, replace the out-of-specification NICs, instead of using the
debounce option. This feature is first integrated into software version 5.3(5)CSX.
For the Catalyst 2900XL and 3500XL, the interface command carrier-delay time can be adjusted
to four seconds as a possible workaround for this same issue.
Refer to Fast Ethernet Consortium Physical Medium Dependent Test Suite for more information
about jitter tolerance.
Performance Notes
Most performance issues are related to switch port configuration, duplex mismatches, link up/down
situations, and data link errors. When you troubleshoot performance issues, review all previous sections of
this document. After you review these sections, proceed to the next section, Understanding Data Link
Errors. The final step in order to resolve any performance issue is to obtain a sniffer trace. A sniffer trace
is very conclusive with regard to any specific performance problem because it details packet transfer.
Understanding Data Link Errors
Many performance issues with NICs can be related to data link errors. Excessive errors usually indicate a
problem. When operating at a half-duplex setting, some data link errors such as FCS, alignment, runts,
and collisions are normal. Generally, a one percent ratio of errors to total traffic is acceptable for half-
duplex connections. If the ratio of errors to input packets is greater than two or three percent, performance
degradation can be noticed.
In half-duplex environments, it is possible for both the switch and the connected device to sense the wire
and transmit at exactly the same time and result in a collision. Collisions can cause runts, FCS, and
alignment errors, caused when the frame is not completely copied to the wire, which results in fragmented
frames.
When operating at full-duplex, FCS, cyclic redundancy checks (CRC), alignment errors, and runt counters
are probably minimal. If the link operates at full-duplex, the collision counter is not active. If the FCS, CRC,
alignment, or runt counters increment, check for a duplex mismatch. Duplex mismatch is a situation in
which the switch operates at full-duplex and the connected device operates at half-duplex, or the other
way around. The result of a duplex mismatch is extremely slow performance, intermittent connectivity, and
loss of connection. Other possible causes of data link errors at full-duplex are bad cables, a faulty switch
port, or NIC software or hardware issues.
When you troubleshoot NIC performance issues, view the output of the show
port mod/port command
and the show mac mod/port command, and note the counter information.
Table 2—Explanation of CatOS show port Command Counters
Counter Description
Alignment
Errors
Alignment errors are a count of the number of
frames received that do not end with an even
number of octets and have a bad CRC.
FCS
FCS error count is the number of frames that
were transmitted or received with a bad
checksum (CRC value) in the Ethernet frame.
These frames are dropped and not propagated
onto other ports.
Xmit-Err
This is an indication that the internal transmit
buffer is full.
Rcv-Err This is an indication that the receive buffer is full.