How many CRCs are an acceptable number?

[JBruyet]

Hey all,

How many CRC errors are considered acceptable? I've heard one a day, one in 5,000, none. I'm having some problems and one of my support techs said that my number of CRC errors is ok for the number of packets that have gone through that port on the switch.

Thanks,

Joe Brouillette

 

[JBruyet]

AND, since there are several error categories:
RX errors
FCS
Alignment
Runts
Giants
TX errors
Drops
Collisions
Late Colln
Excessive Colln
Deferred
I'd like to know what numbers of errors are "acceptable" for each category. I've tried to contact HP's support but I haven't heard back for a while. Any help would be greatly appreciated.

Thanks,

Joe Brouillette

 

[ralpheb]

Hello Joe -

There are various guidelines for acceptable bit error rates. For example, the IEEE has a guideline. I do not remember what it is, but I seem to recall that it is much lower than 1 in 5,000. I think it is more like 1 in a billion.

The IEEE guideline is oriented toward what a very healthy LAN should be able to attain on a cable. In most real world networks, there are higher levels of errors which don't matter much, and so few network administrators go to the bother of trying to achieve the error level of the IEEE guideline.

I would be surprised to learn that the IEEE guideline takes into account the possibility of duplex mismatches. For this reason and the reason I gave in the previous paragraph, others recommend much more liberal bit error rates.

At

http://www.cisco.com/warp/public/473/46.html

Cisco points out that you should expect a lower error rate on full-duplex links than on half-duplex. They go on to say that an error rate of one percent is acceptable and that at an error rate of 2-3 percent there may be a noticeable performance degradation.


Hewlett-Packard has a document at

http://www.hp.com/rnd/library/troubleshoot_lan.htm

that covers somewhat similar ground to the Cisco document I referenced above. H-P's rule of thumb is to aim for no more than one error packet in 5,000.

Why are Cisco's recommendations and H-P's different? The error rate at which the user will notice performance degradation depends very much upon the nature of the traffic. Factors such as Transport window size, retry latency, clustering of bit errors (that is, are they bunched together or spread out?), and end-to-end latency determine the extent to which a particular bit error rate will result in performance degradation. Cisco and H-P are likely modeling their math on somewhat different traffic assumptions.

A very simple and easy approach is to let the ProCurve Fault Finder (look for "FFI" entries in the Event Log) feature tell you whether you have too many errors or not. You can adjust the Fault Finder sensitivity based on whether you would like to be aggressive or not in achieving a very clean LAN.

Regards,
Ralph

 

[BAsh12]

Basically none. What is a CRC is what you have to ask.

You may see the occaisional one on a good network,
Unpluging the host and plugging it back in whilst it is live is most common. You can prove it. Example. Get on the console port and do "clear statistics" a "sh int A1" and you should see counters all at zero. Unplug the host and plug it back in, you will probably see a CRC.

The main problem is normally one side working at full duplex with the other at half. Normally caused by someone hardcoding one end and leaving the other at auto-negotiate. This is very very bad, you will see 50% or above CRC for the port.

It should never happen on a full duplex link as it will no longer be CSMA/CD. If you do, you have a problem. Start with the cabling. Is the patch lead ok? Are all termination connections correct? Is the cable infrastructure CAT5/5E/6 certified. That is, fully tested?
If it is a cable problem, you will also see FCS, short frames etc.

Once you have ruled this out, check the drivers for the network card, a very common problem. There is also a Microsoft known issue 315237, Inter frame gap. If it is a driver problem, you would normally see a similar amount of FCS to CRC.

When all is checked and the problem still exists, try another network card.

So in short, on a switched network running F/D, it's bad.

Call me a perfectionist.