There are several things to watch for in relation to collisions:

- Does the detected collision level track approximately with the utilization
level?
If changes in utilization and collision levels track together
reasonably closely, then there may simply be too many stations transmitting on
the collision domain, assuming that there is a collision problem at all.
- Are there spikes of detected collisions that do not follow the utilization
level?
The collisions may often be traced back to a single source. It may be a
bad cable to a single station, a bad uplink cable on a hub or port on a hub, or
a link that is exposed to external electrical noise. Over time the problem
station may be isolated by monitoring the traffic sources at the same time as
the collision level bursts higher. If the problem seems to relate to
transmissions from a single station then troubleshoot that suspect link. If the
problem seems to relate to transmissions from multiple stations then compare
that information against the functional network diagram to see if there is a
common path between those stations and the rest of the collision domain. The
single station could be an uplink from the collision domain to a switch, and
the functional diagram should reveal that all of the other stations are beyond
that link. One or more stations set to full duplex within a collision domain
will also cause this sort of collision problem, as well as other errors.
- Are there collisions when there is no apparent utilization to cause them?
It could be a noise source near a cable or hub. Use divide and conquer
troubleshooting to isolate the location of the fault, adding traffic to the
network from the monitoring tool while troubleshooting. This sort of fault must
usually be diagnosed after-hours since portions of the collision domain will be
disconnected from the network during troubleshooting.
- Are there approximately 33 percent or 100 percent collisions?
There may be a 100 Mbps station attempting to connect to a 10 Mbps
segment. This collision level results from a station transmitting MLT-3 encoded
100 Mbps signal to the 10 Mbps hub. The 10 Mbps hub turns on the link state LED
and forwards its best interpretation of the MLT-3 signal as Manchester encoded
data. The 100 Mbps end is not able to establish synchronization, and thus does
not turn on the link state LED or forward any received traffic to the MAC
Layer. The reverse situation does not result in a problem. If a 10 Mbps station
attempts to insert into a 100 Mbps only hub, then that station will not achieve
MLT-3 synchronization and the hub will not turn on the link state LED or
attempt to interpret and forward the Manchester encoded signal.
To track
down the source of collisions, it is often necessary to have traffic on the
network. Use a traffic generator to add a small amount of traffic while
monitoring. A safely insignificant level of traffic is 100 frames-per-second,
100-byte frames, which is still sufficient to sensitize nearby faults and allow
them to be located. Using a destination address within the collision domain
will prevent the traffic from crossing bridged connections and disrupting other
users. Do not make up a non-existent destination MAC address because that will
flood to all parts of the broadcast domain. If the generated level of traffic
is low, then the destination address could be set to that of the tool or
station generating the traffic without disrupting its operation.
Some
media-related problems are traffic-level dependent. Try gradually raising the
traffic level to more than 50 percent, and at the same time watching the error
and collision levels. Many monitoring tools offer LED indicators for both,
which makes it much easier to vary the traffic level while watching for
resulting errors or elevated collision levels. Solving collision-related
problems can be tricky because the measurements are largely dependent upon the
observation point. Results may vary between two observation points separated by
only a few feet on the same cable. Make tests from multiple locations and watch
for changes in the nature of the problem.
If collisions get worse in
direct proportion to the level of traffic, if the amount of collisions
approaches 100 percent, or if there is no good traffic at all, the cable system
may have failed.