IP device tracking on 3850 Switches

[Jason Flory]

Hello,

We have 2 3850 stacks in our corporate office.  One stack is for users and the other is for severs.  We have been having alot of false positive duplicate IPs errors on our servers and have been find many others out there having issues with IPDT on 3850s.  The recommendation is to disable.  I want first fully understand this and second make sure I am not creating other issues down the line.

In our environment the server stack is the one that has the SVIs and does all layer 3.  We also have several other stand alone switches that connect our blade servers to the server stack.  

First question; in a multi switch environment which switch manages IP Device Tracking?  When I do show IP device tracking all on each switch it seems that a lot of the IPs being tracked are not directly connected nor are they only on the layer 3 switch.  Is there a command to tell the switch who should be managing IPs?  

Second question; if we turn this off using nmsp attach suppress on each interface what are the implications?

[jnewb2015]

Having the same issue here. Simply changing the ip device tracking interval to 10 on our 3850 stack caused an issue. All clients on the network went down for at least 5 minutes. I can't tell you what the implication are to using the nmsp attach suppress but I can speak to the fact that changing the interval caused an issue. All work relating to this feature should be done outside of production hours. 

[Jason Flory]

I think that is  the probe delay from 20 to 10.   I played around with that but did not cause any additional issues but did not fix anything.  From what I understand you cannot turn off IPDT globally.  The nmsp attach suppress  here is supposed to turn off completely instead of changing delay.

[jnewb2015]

Thanks for the reply. The reason I changed it from the zero default value to 10 was on the recommendation of this document: 

http://www.cisco.com/c/en/us/support/docs/ios-nx-os-software/8021x/116529-problemsolution-product-00.html

I was very surprised to have the switch apparently clear arp tables. There was no downtime on the switch itself however, just my network. Since the value has been at 10 I have still been seeing DHCP issues but they appear to be less in number. This may just be coincidence as I do agree that turning the feature off completely is in fact the solution even though the link I referenced above hints at a delay of 10 being plenty of time. 

I should also note because it may of some interest here that the only devices (with the exception of one PC) we have on the 3850 stack are IP Phones. During this 5-7 minute outage the phones did go down as one would expect if there was an issue with the 3850s. However, most clients connected to our 3750 stack lost connectivity temporarily also (PC's, printers, ect). These are two separate stacks and if the command cleared the arp tables on the 3850's the only devices that should have had issues would have been the phones.