Collisions....What is the standard???

stevennowell · ‎04-22-2007

I am finding conflicting info among cisco documentation for the acceptable threshold of collisions on a ethernet segment. A CCDA question was that the Cisco standard for collisions was .1% of total throughput. Divide collisions by packet output to arrive at this number. Assuming that there is no collisions on full duplex, this must therfore apply to 1/2 duplex operation.

CCO on the issue says that "collisions do not constitute a problem". "In conclusion, the collisions counter does not provide a very useful statistic to analyze network performance or problems. " http://www.cisco.com/warp/public/63/eth_collisions.html.

The former seems like the most logical, but I have interfaces in the network I support clocking 12% collisions per the equation above, and there is not users complaining of latency on these segments. I have seen interfaces communicating with a duplexing mismatch go above 30% collisions and it is my belief that the 12% segments are set correctly. I should also add that these segments are connected to a baystack switch, so i think this could be the cause.

What is the standard that you use for this?

ariela · ‎04-22-2007

Hi,

the CCDA is the theory, the CCO is more a best practice.

A good guide, in theory, to use is collisions should not total more than 1% of the frames transmitted. The following table gives a guide as to recommendations of collision levels with respect to bandwidth utilisation of a segment:

%age Utilisation/Maximum %age Collisions

less than 20% / 1%

20-49% / 5%

over 50% / 15%

Remember that, in practice, collision do not constitute "always" a problem. When I have collisions, as CRC or late collisions or errors in general, normally I check my topology, essentially L1 and L2.

HTH

Andrea