3850 Switch with DHCP odd behavior

[ken.montgomery]

We have begun using the new generation of 3850 switches, specifically WS-C3850-24P-P switches.  We use the switch to supply DHCP for the VLANs. 

We are running software version:

cat3k_caa-universalk9, version 03.03.00SE

So, we built the pool according to these guidelines:

ip dhcp excluded-address 10.96.81.200 10.96.81.209

ip dhcp excluded-address 10.96.81.220 10.96.81.254

ip dhcp excluded-address 10.96.81.1 10.96.81.12

ip dhcp excluded-address 10.96.81.30 10.96.81.50

!

!

ip dhcp pool LEGACY-DATA

network 10.96.81.0 255.255.255.0

domain-name eng-prod.com

dns-server 10.1.1.10 10.96.200.11

default-router 10.96.81.4

So, here is the kicker... users are not getting addresses.  The dhcp server doesn't respond BUT... it responds to a certain point.  Immediately after restarting the DHCP service on teh switch, this is what the pool statistics show:

Pool LEGACY-DATA :

Utilization mark (high/low)    : 100 / 0

Subnet size (first/next)       : 0 / 0

Total addresses                : 254

Leased addresses               : 0

Excluded addresses             : 148

Pending event                  : none

1 subnet is currently in the pool :

Current index        IP address range                    Leased/Excluded/Total

10.96.81.13          10.96.81.1       - 10.96.81.254      0     / 148   / 254

According to my math, there are 54 exclusions.  The pool shows 148 before any leases handed out, and as each lease is handed out, the excluded rises.  I know the older version of this new Cat-os had DHCP issues... does this version as well?

Any suggestions?  Our pool appears to run out of leases iwth about 100 leased, showing 106 leased, 148 excluded, for a total of 254... which sounds right, but there are only 54 excluded.

Help?

[ken.montgomery]

1. If the VLAN was out of scope, addresses would not be handed out, which is happening successfully.

2. If VLAN was shut down, addresses would not be handedout.

Addresses are being handed out.  Just not enough of them.

It is not a layer 2 issue.

[paul driver]

Hello

"Again, that shortens the lease, but doesn't solve the problem.
When my office is filled with staff, there still won't be enough addresses, as the system isn't handing out the full range"

A good possibility is due to the high dhcp lease of 2 days and if you have mobile clients this would make exhaustion even more so.

I would suggest to shorten the lease time to a low value (maybe 3hours given the size of your scope it looks like you could take the extra BW) and clear the binding database again and force the clients to renew also (ip config /release/renew)

If this resolve your issue then increase the scope back to a more applicable level

Res
Paul

[ken.montgomery]

Ok, maybe what I'm telling you isn't sinking in.

There are probably 100 clients in the office.  These clients get one address each.  Oftentimes, all 100 will be in the office at once.  When they are all there, it doesn't matter if the lease is 5 minutes or 5 days, as they are all there.  At those times, 100 addresses would be needed.  Once we get above a certain level (say 70), there are no more addresses available.  (These are just figurative numbers, I'd need to go back to figure out the exact ones, something like 106 or 108 is the maximum being handed out).

The lease time will have no affect on this.  As the problem occurs when everyone is in the office, not when everyone transitions in and out. 

The problem is the DHCP server on the switch doesn't believe it has enough addresses to hand out, as the excluded number is way too high. 

[dkeenum]

I know this is a very old thread, I ran across it today because I was having the same problem with a 3850 that I run DHCP on.

I have a 20 address pool to hand out IPs for our WAPs.  All the WAPs went down after a circuit outage and after doing some research I noticed that my pool said that all IPs were excluded.

Pool WAPs :
 Utilization mark (high/low)    : 100 / 0
 Subnet size (first/next)       : 0 / 0
 Total addresses                : 254
 Leased addresses               : 0
 Excluded addresses             : 254
 Pending event                  : none
 1 subnet is currently in the pool :
 Current index     IP address range                    Leased/Excluded/Total
 0.0.0.0           10.48.38.1       - 10.48.38.254      0 / 254 / 254 

Turns out that the switch put all the IPs on to the ip dhcp conflict list. (don't know if this was caused by the outage, still researching) According to some documentation I read, once an IP is on the conflict list it gets excluded from the pool and will not be available again until an admin manually clears it with "clear ip dhcp conflict *".  Once I did this my pool returned to the way it was supposed to be and my WAPs came back up.

Pool WAPs :
 Utilization mark (high/low)    : 100 / 0
 Subnet size (first/next)       : 0 / 0
 Total addresses                : 254
 Leased addresses               : 18
 Excluded addresses             : 234
 Pending event                  : none
 1 subnet is currently in the pool :
 Current index     IP address range                    Leased/Excluded/Total
 10.48.38.38       10.48.38.1       - 10.48.38.254      18 / 234 / 254 

I don't know if this helps, but, I figured I would post just incase anyone else stumbles across this thread like I did.

-David

[edondurguti]

I have the same problem

[airfarespare]

Thank you David, this helped me solve an issue that I've had for over 6 months !

I also found this, RE: Auto clearing the conflict list... https://community.cisco.com/t5/network-management/auto-clear-ip-dhcp-conflict/m-p/1530142/highlight/true#M73363

Thad