When monitoring interface errors, what percentage of traffic should you set your 'critical' threshold to according to best practices and does it depend on the interface type (T1, Ethernet etc)? It would be a huge bonus if you can explain the justification for the particular percentage. I've found a few thread comments on various sites that mention 1%, but with no real justification.
Ethernet standard officially allows 10^-12 bit-error-rate, while in practice the hardware meet much better BER than which standard demands.
You should also be able to bing for 'SQA' (Service Quality Assurance) or 'SLA' (Service Level Agreement), some companies publish them, you could use them to check what your competitors are offering and offer something to that level.
Our SQA states to customers that 0.02% is minor fault (we will fix if ticket is opened), which I think is quite large packet loss for fibre connection, but same SQA covers also DSL so we didn't want to be too aggressive with it. So far this has been sufficient to customers, but we are prepared to reduce the number if it is hurting sales.
There are several bingable tools online, where you can check how much packet loss hurts TCP, which can be useful information when deciding what is acceptable loss for your application/product:
How long to generate traffic?
Ethernet/SONET networks are often deployed with GBICs,SFPs,or XFPs.Depending on the quality of the GBICs,SFPs,and XFPs, the network may havea guaranteed error rate of 10-12 to 10-15. For a standard GigE network at an error rate of 10-12, the user will experience a few errors a day. At an error rate of 10-15, the user will experience about one error a week. A 15-minute error-free test should be run to confirm that the network is properly provisioned and that the circuit is capable of handling the traffic. Longer tests offer more statistical information into the quality of the circuit and shorter tests offer less information.
For 10-Gigabit Ethernet there is a BER recommendation of 10^-13 to acheive the same number of errors per day as for Gigabit Ethernet.
Some other sources say 10^-12 for 10GE. If you set it to both 10^-12 for both GE and 10GE that would probably be a good number.
1% loss on a 10g link can be a huge problem these days for many network devices. If you're looking at inside a rack / datacenter, you should really be trying to minimize the instances of packet loss / errors within your network.
Errors can also be things like switches turning into hubs by accident (Dell 55xx switches have a bug in them in some firmware versions that disables their mac learning after 49.5 days) - One reason we caught it was because we had been tracking the links over a long time and got flooded with traffic for MAC's we didn't have on our switch.
Also for those with AC or other lines installed in a tray, you might see some increased error rates, which might help prove to management that you should really re-locate some of the cable in the plenum or upgrade to fiber lines for these runs.