6500 Supervisor Failure with DFC Enabled Line Cards

Unanswered Question

Setup


6509-E w/ Dual Sup-720


2 Line Cards - 6748s w/ DFC installed


Assume that the CEF tables are synchronized between the DFC and the Supervisors and that NSF and SSO are both enabled.


Question: Both Supervisors Fail - How long could the DFC CEF tables continue to forward traffic before being invalidated? Layer 3 traffic? Layer 2 Traffic?


Since there is no supervisor involved here, what about MAC/CEF aging? Would it take place on the DFCs or would they just keep their caches indefinately until the control plane returned to operation?



  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 5 (2 ratings)
Loading.
Edison Ortiz Wed, 08/19/2009 - 05:42
User Badges:
  • Super Bronze, 10000 points or more
  • Hall of Fame,

    Founding Member

The answer depends upon the type of failure. If Both Supervisors fail, most likely the reason is due to the switch fabric which will also affect any DFC to DFC forwarding decision between DFC enabled modules.


Now, let's see the switch fabric is fine - the DFC will maintain the data plane information but not the control plane information so L3 traffic will automatically be down without the Supervisors but some L2 traffic may be working if the forwarding lookup from the source to destination is within the DFC at this moment.


This is a corner case scenario and you may encounter different results every time you try it.


HTH,


__


Edison.


Giuseppe Larosa Wed, 08/19/2009 - 06:33
User Badges:
  • Super Silver, 17500 points or more
  • Hall of Fame,

    Founding Member

Hello Michael,

DFCs can receive a copy of the CEF table and other data from master copy on PFC of a supervisor.


They are "consumers" of CEF data and a PFC is a producer of CEF data.


What they can do without a running supervisor feeding them is to use the current table until they detect no supervisor is available.


I think it is a question of implementation they could stop to work after a reasonable timeout.


There is no one that can react to a change that can make an ARP requests.


I would expect the system to die in 2 to 4 of minutes but I've never done tests on this.


Hope to help

Giuseppe


Actions

This Discussion