s-mlt and high cpu utlization on 8600

[jlkilpat]

Existing network is passport 8600 as L2 switch with gig ethernet to IDFs. A 2nd 8600 was added for S-MLT. Now, configuration is two 8600s configured for s-mlt with interswitch trunk link on gig ethernet 1/1 and 2/1.
20 wiring closets are primarily 450t with some 5510.
Passports are running 3.5.1 code, 450T and 5510 are running latest code from Nortel's web site (4.5.2 and 4.2.1). 8600s are configured with default VLAN and have a new port based VLAN for each closet added as MLT is introduced into the network over time. Each VLAN has it's own IP address and VRRP is configured. The gig ports are setup for NO autonegoiation; tagging; discard untagged frames. 6 closets worked fine, then another 5 closets were added and everything worked fine for about 8 hours. Then the Master 8600 began experiencing 100% CPU utilization and two closets went down. These closets were added that day and both consist of a stack of 5 450Ts. This occurred 2 to 3 times over the next several hours; when the closets would go down, the 450T stack would NOT show a unit 1 in the stack. The 8600 CPU utilization would eventually go back to near zero; rebooting the 450T stack would make unit 1 reappear; though the stack may not be seen until the fiber cable was reseated. Nortel support looked at the config files and could not find any problems. I did reboot the master switch and it seems that I've not seen the 100% cpu utiilization since. I can't say for sure if the closets went down and CPU utilization up or if the CPU utilization killed the closets. There has not been any issue with the 5510 closets.

My questions are:

anyone else seen this?

Anyone running S-MLT with at least 2 dozen MLT connections over gig ethernet?

Is the 450T really capable of running S-MLT and working with Gigabit ethernet over fiber. There is a lot of info in the release notes about this. Is version 4.5.2 code for the 450 T stable?

Do I need to upgrade the 450Ts?

 

[atonns]

There is a running issue for HA and SMLT on the 8600s and the 3.7.x line of code. I'm running 3.7.4 and have disabled HA because of this reported instability. It was supposed to be fixed in 3.7.8, 3.7.9 and 3.7.10 but it's still "being worked on" accoring to this bulletin:

http://www130.nortelnetworks.com/cgi-bin/eserv/cs/main.jsp?cscat=DOCDETAIL&DocumentOID=367129

I'd double-check the bulletins for all the specifics for your setup, as it sounds REALLY similar to the problem above.

Mind you, I'm not actually experiencing this problem - but am aware that it exists.

Tony

 

[Jynxx]

What do your spanning tree configs look like?

Also...in regards to 450's and SMLT...any switch...by any manufacturer that I am aware of can work with an SMLT because according to the 450 it is just an MLT...it's the 8600's that are configured for the SMLT with the IST...

 

[jlkilpat]

Spanning tree configs are set to disable.

 

[ecuamer]

Have you look into the log files when the CPU utilizaton increases ( show log file tail ).
The edge switches ( 450 or 5510 ) as mentioned only are capable of mlt as smlt is only configurable on the Passports.
1. STP is enable globally but should not be enabled on any of the fiber connections to the closet that are running smlt. IST stp is disable as well.
2. Are you running any kind of multicast traffic ? IGMP snooping should be enabled on the closet switches if you are running multicast traffic ( PIM,DVMRP) .
3. Never ran 3.5.1 only 3.5.2 and that was pretty stable .Currently being forced to upgrade to the 4.0.2 code.