06-12-2017 05:14 AM - edited 03-08-2019 10:56 AM
Hey all!
I read some topics but didn't really find an answer.
Why do I get output errors on almost all interfaces of all my 3850?
I get no input errors, CRC errors, etc.
Only output errors.
And there aren't any problems. Cabeling is brand new.
Is that a bug?
PS: version 03.06.06E
06-12-2017 05:49 AM
3.6.6 is pretty stable not ruling out a bug but its definitely one of the better versions currently
it may not be the switch at all either it could be burst traffic from application traffic temporarily flooding the port filling the buffer up so it cant hold the traffic and has to drop it , interface could be overloaded as well check what its pushing on average when you see the drops what's the tx and rx levels running at
whats the ratio of traffic in total going out compared to the output drops counter itself , if its less than 1% it could be legitimate aswell anything over that its advised to check it
identifying the traffic that's being dropped would be a first step id try using something like splunk or wireshark to capture it and see exactly what it is, if its the same type of traffic constantly it may point you to something
https://notalwaysthenetwork.com/2014/01/06/microburst-detection-with-wireshark/
do you have any qos enabled at layer 2
unfortunately there no 1 answer for output drops it can be a number of things that can cause it and you have to work through it to find the root cause
06-12-2017 08:10 AM
06-12-2017 08:14 AM
huge amount of output errors on it , clear the counters and monitor it until the morning
clear counters gx/x
see if those errors increment with the outputs , if they are your over utilizing the switch port too much traffic coming in
whats the full interface output has reliability dropped below 255 ?
if it is change the local cable too
06-12-2017 08:18 AM
06-12-2017 08:25 AM
hmmm what the tx/rx and reliability show you , reliability should be 255
have you tried changing one cable on one effected device , thats a huge amount of errors in 1 days monitoring
how much traffic are the workstations pushing through the interface
reliability 255/255, txload 1/255, rxload 1/255
these are standard copper connections yes , check the tdr cabling too its a basic check to show cabling from switch point of view
https://supportforums.cisco.com/document/74231/how-use-time-domain-reflectometer-tdr
im running 3.6.6 globally on lot of 3850s without seeing these issues with workstations
06-13-2017 02:57 AM
06-13-2017 03:18 AM
hmm that's odd because pause frames come in when port is overloaded and same when output errors and total output drops increment together , pause frames are a form of flow control for the switch port
you could try change the image but I would wirehsark the interface first find out what type of traffic is really going into the port as what the switch is showing and what your saying the notepad is only doing don't really match , because the interface at times is being overloaded and its showing that with these types of errors incrementing
do you manage your switches with any software that does history of the interface tjat you can measure the traffic or look back at ,splunk , prtg , solarwinds that can help show the history of the port traffic
I was suggesting to change the cable as a test to be sure on one device just to be sure its not that
your last port wasn't showing pause frames you posted and was recently cleared too , that ports is running length of time of the swiotch are the pause frames currently incrementing
You have couple of choice here , actually find out at the wire level what the traffic is then tray and fix it with qos or other features or upgrade is you think its software issue but you could be back to square one if its not
07-03-2017 01:42 AM
Hey Mark,
I stumbled over this thread:
https://networkengineering.stackexchange.com/questions/29103/output-drops-errors-on-cisco-3850
"Common Cause: This issue is due to the low Output Queue size."
Can I increase the output queue size?
The current size is 40.
In comparison, the max input size is 2000!
07-03-2017 02:14 AM
You can but beware increasing it may cause latency to go up to too so I would test it bit by bit and not just open it up
07-03-2017 02:16 AM
How?
07-03-2017 02:29 AM
as packets could end up in the queue longer before being sent
07-04-2017 03:21 PM
Post the complete output to the command "sh interface <BLAH> cont". And no, do not post the screenshot of the output.
06-13-2017 04:42 AM
Hi,
I have seen this in a number of 3850. Sounds like a bug. The following are a number of bugs that are out there:
CSCur94225
CSCvb65304
Thanks
John
06-13-2017 05:17 AM
oh okay!
but what would be a solution?
3.6.6 is the recommended image from cisco.
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide