• September 24, 2020, 07:30:07 AM
Welcome, Guest. Please login or register. Registration is free.
Did you miss your activation email?

Author Topic: Loop  (Read 2744 times)

0 Members and 1 Guest are viewing this topic.

Offline ming_z

  • Rookie
  • **
  • Posts: 5
Loop
« on: June 30, 2015, 10:37:08 AM »
Itís a typical network: all edge switches and 4 wireless controllers are connected to two 8600s (5.1.8.3) through SMLT. It started with some error messages first (seems related to multicasting). Logged a case with Support. Nobody was taking them seriously because there was no impact. Suddenly CPUs on both 8600s jumped to 99-100%. They would go down then up... One core stopped forwarding some trafficÖ Finally wireless controllers were identified as the culprits. Fdb entries from 5 minutes apart on the same switch were compared and many MAC addresses were learnt via one port but via a different port minutes later. All MAC are wireless clients and all ports involved are connected to WLAN controllers. I argued this might be caused by roaming Ė clients roam between controllers. But after I disabled all involved ports on one 8600, CPUs came to normal immediately. Even those irrelevant errors messages disappeared right away. The controllers are the old Nortel WLAN 2382 (rebranded from Trapeze). Two SFP ports on the 2382s are configured as Port Group and this setup has been working for years without any problems. Checked the config/logs and didnít see any problems. Now I guess I cannot trust the Port Group anymore. Has anyone seen this type of setup/problem?


Offline Telair

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 965
Re: Loop
« Reply #1 on: July 02, 2015, 03:44:39 PM »
We had issues with Sun systems broadcasting the same MAC address on multiple NIC's and yes, it really can drive up CPU utilization quickly.  Are the controllers connected via SMLT's or LACP LAG's?  Are they connecting directly back to the core or with a switch or something in between them?  Sorry, I never used the old Nortel controllers, so just going over the basics to start.

Offline ming_z

  • Rookie
  • **
  • Posts: 5
Re: Loop
« Reply #2 on: July 02, 2015, 03:59:31 PM »
The controllers are connected to core switches directly through SMLT. I just donít understand why so many MACs were learnt via one controller but re-learnt via another controller in 5 minutes. It sounds exactly like roaming but apparently itís not.
After reading the following post, I am thinking to use LACP with SMLT on 8600s to connect those controllers because Trapezeís Port Group should be LACP:
blog.michaelfmcnamara.com/2009/08/lacp-configuration-examples-part-2/

Offline Telair

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 965
Re: Loop
« Reply #3 on: July 02, 2015, 04:13:44 PM »
Yeah, since those units were Trapeze units just re-branded I wouldn't expect them to understand MLT's and would go with using LACP LAG's instead as long as the 2382's support it. 

Otherwise maybe consider connecting the controllers to a switch which then connects to the core?  The switch can deal with the MAC's jumping around from port to port and it just forwards down to the core over a couple SMLT links.  So the core just sees the MAC's coming down over the same links and not all over.

Offline ming_z

  • Rookie
  • **
  • Posts: 5
Re: Loop
« Reply #4 on: July 02, 2015, 05:07:31 PM »
Thanks Telair. I am actually thinking the second options as well. I just don't have enough 1G ports where those controllers reside for now.

Offline TankII

  • Hero Member
  • *****
  • Posts: 556
Re: Loop
« Reply #5 on: September 21, 2015, 10:45:12 AM »
This was a code bug in earlier versions of Trapeze/Nortel code, and was addressed in later versions.
We upgraded our MX400's to Trapeze code (while in SMLT mode).  Eventually we replaced them with MX800's.
TankII