Catalyst 3850 High Memory Used

[jluis_alves]

Hi,

We have some catalyst 3850 switchs with more than 80% Memory usage.

Looking on memory allocated we can see that arp input have highest memory usage.

show processes memory sorted allocated
Processor Pool Total: 886455648 Used: 364614224 Free: 521841424
lsmpi_io Pool Total: 6295128 Used: 6294296 Free: 832

PID TTY Allocated Freed Holding Getbufs Retbufs Process
35 0 219477763952 219476780384 1037120 0 0 ARP Input
527 0 103328177000 111161034504 123504 0 0 SNMP ENGINE
526 0 60720782024 53033593312 119096 0 0 PDU DISPATCHER
10 0 43242937440 43280730664 1684264 41208919058 41212352574 Pool Manager
520 0 28481858096 28481857952 69144 0 0 mdns Timer Proce
191 0 17112531160 7484608760 89464 0 0 CDP Protocol
203 0 15545194768 9467824 72208 61600 0 IP Input
0 0 11601240976 11249616384 5052280 17486679 0 *Dead*
117 0 10562472296 10241322360 636536 1474704 0 IOSXE-RP Punt Se
36 0 10348925360 10349181064 45000 0 0 ARP Background

Is this behavior normal?

[jasonm002]

If you can't run 3.6.8 (for example, because you have mgig/NBASE-T 3850s), try running 16.3.7 on it if you haven't already, especially if you find the switch freezes when you do a "show run" or a "wr mem".  16.3.6 or earlier for 3850 had some catastrophic bugs such as CSCvj49476, which we ran into.

Also you will want to apply a device-tracking profile similar to the following to your uplink ports if you have enabled DHCP snooping in 16.x, because DHCP snooping in 16.x creates device-tracking profiles which you cannot remove for every VLAN you enable it on. You can see these by issuing a "show device-tracking policies" from the exec mode.

Global:

device-tracking policy DT-NOTRACK
 trusted-port        
 security-level glean    
 device-role switch    
 no protocol ndp    
 no protocol dhcp6    
 no protocol arp    
 no protocol dhcp4    
 no protocol udp    
 tracking disable
 
! Switch uplink interface config:
device-tracking attach-policy DT-NOTRACK