Hi every body!
please consider the following scenario:
sw1 (halfuplex)--------(full duplex) sw2
My book says in order to determine duplex mismatch, watch for increase to certain counters on half duplex interfaces. The counteers such as runt,collisions and late collisions occur when the other device uses full duplex.Note these three counters can also increment when legitimate collisions occur as well."
Since the reason for these counters to increment could be because link is operating in half duplex or the there is duplex mismatch . We can not specifically look at these counters and conclude since thse counters are incrementing there must be duplex mismatch. It is well possible that link is operating in half duplex in that case it is normal for these counters to increment.
Am i correct?
Thanks a lot !
". . . 0.1 percent . . ."
Is this correct? (I.e. possible typo?)
Been a while since I've worked with production half-duplex Ethernet, but recall collision rate was higher if more than one host offered high load at the same time. Following one of the references from your reference (http://www.ethermanage.com/ethernet/pdf/techrept13.pdf), sections 4.x discuss this. (I.e. it's normal [high collisions] for high offered load situations, and doesn't usually adversely impact performance.)
Reason I ask, if collision counter can be high for some "normal" network situations, it wouldn't then always be suitable for detecting duplex mismatch problems.
[edit - thinking some more about this, I believe the overall collison rate would likely be a low percentage over longer time periods although might be very high across a short time period. Something like a duplex mismatch could, I expect, make for a much higher long term collision percentage. As I noted, it has been a while, and I now recall I never usually took much interest in collision percentages, more so interested in overall load utilization.]
As for the original question, I don't recall all the interface counters I look at. Perhaps best, if possible, to confirm active duplex mode being used on both ends if there's a question of both using the same mode. However, recall that other "unusual" error counts, e.g. (and especially?) CRC errors give a clue that there's something very wrong. Well that, and performance much, much worse than nominal bandwidth.
"So if i find late collision, can i safely conclude there is duplex mismatch ?"
Not necessarily although it is a good indicator. If i see late collisions i usually think of duplex mismatch but late collisions can also be caused by a faulty NIC for example.
in an half-duplex both sides scenario collision rate for a healthy link is less then 0.1 percent in comparison to successfully sent packets.
Late collisions should not be present.
It is likely that if you have a duplex mismatch the rate of collsions is bigger and you see increasing late collisions.
In practice as Glen notes CDP fills the log buffer with warning messages about duplex mismatch.
The effects of duplex mismatch becomes important when traffic volume on the link increases.
Hope to help
Yes that is correct if the other end is also half duplex you would also have runts collisions and possibly late collisions . That being said there are few devices that work at half duplex anymore. Also between cisco devices cdp will tell you in the logs that you have a duplex mismatch if one is present .