Skip to content. | Skip to navigation

Personal tools

Navigation

You are here: Home / Wiki / Kb230

Kb230

Emulab FAQ: Testbed Operations: Problems with switches dropping out of the switch stack.

Emulab FAQ: Testbed Operations: Problems with switches dropping out of the switch stack.

> Our Nortel switch stack has 7 pizza boxes.
> Since the weekend, some of the boxes randomly drop out of the switch  
> stack.
> Power cycling puts the units back to the stack, but it does not last  
> more than half a day.
> Then, some other units drop out next...

I would recommend looking to see if there were any experiments that were swapped in every time the switch failed. In our experience, it's not too uncommon to have experimenters send traffic that makes the switches behave badly.


Update from Keith Sklower, 10/04/06 16:26

What the problem actually turned out to be was that I was trying to enable multicast snooping via SNMP, and this is on not just the slow path but the slow-as-molasses-in-january-slow path, so the snmpit code was retrying much too agressively, and the nortels seemed to regard the retried requests as new ones, got behind, got mad, crashed, etc...

And, IGMP snooping is unnecessary anyway. The current nortel snmpit module now, after it does any potentially challenge set, does a suitably delayed get, retrying that a few times before giving up (for something on the order of 30 seconds) before retrying the set.

IGMP snooping is normally turned off, but can be enabled (at your peril) by setting $self->{SKIPIGMP} = 0; up in sub new($$$;) of snmpit_nortel.pm