Detecting Inactive Neighbors Over OSPF Demand Circuits

1
49th IETF Meeting

• Detecting Inactive Neighbors Over OSPF Demand
  Circuits
• (draft-ietf-ospf-dc-00.txt)
• Sira Panduranga Rao, University of Texas at
  Arlington
• Alex Zinin, Cisco Systems

2
Problem Definition

• Whats the problem
• Hello Suppression when link brought up
• Unable to detect Inactive Neighbors (say
  OSPF process has died on neighbor)
• Problem aggravated in case of Oversubscription
• Observed on point-to-point (p-t-p) links
• Can be extended to point-to-multipoint
  (p-t-mp) interfaces
• When seen
• Topology of Network does not help routers
  (neighboring this Inactive router) communicate
  their knowledge through flooding to other routers
• What Happens
• Routing Table entries always present
• Application traffic sent to the router is dropped

3
Solution

• Techniques (Use either or both of the above)
• Send Hellos when Link is brought up (Hello
  Probing)
• Limit number of LSA retransmits
• Hello Probing
• Need to advertise capability (backward
  compatibility)
• Send hellos when link is brought up and at
  configurable intervals
• Note Hellos should NOT keep the link up
• In oversubscription case, send hellos only to
  neighbors for which a circuit has been
  established
• Routers can probe their neighbors asynchronously,
  do not have to coordinate when Hello and Dead
  timers are started

4
Hello Probing Details

• Hello probing capability Use bit 2 (value 0x04)
  in extended options LLS TLV in DBDs
• Use the same bit in Hello packets as
  Reply/Request (RR) bit. This avoids looping of
  hello replies
• Send Hello probes (Hellos with RR bit set) when
  link is brought up and periodically if circuit is
  still in active state
• Start Dead timer when sending probes
• Reply back to Hello probes with unicast Hellos
  with RR clear
• Stop sending probes when a Hello is received
• Generate 1-WayRecived if Inactivity timer expires
• If interface is p-t-mp, then account only for
  neighbor specific timers

5
Limiting LSA Retransmits

• Count the number of LSA retransmit for neighbors
• Execute a KillNbr or 1-WayReceived event if
  counter exceeds a configured value
• An LSA acknowledgement sets count to zero
• No cooperation required between routers across
  demand circuit
• Implemented and deployed since 1996
• Original idea by Derek Yeung and Padma
  Pillay-Esnault
• Implemented in Zebra DC code lately

6
Also Recommended

• If OSPF process is going down, routers should
  flush its LSAs (or a subset of them)
• (This suggests that routers should look
  for alternative routes)
• Problem with the above is that LSAs cannot get
  through when
• There is loss of packets or
• Demand circuit is down or
• Router is oversubscribed

7
Considerations

• LSA retransmit limit does not require the other
  side to cooperate
• Hello Probing is backward compatible through
  capability announcement (use of HP bit in
  extended options TLV)
• Added cost more data sent on link

8
Discussion

• Agree to use a bit in the standard Options field.
  Makes implementation very simple and will not
  require any additional mechanisms to fix the DC
  spec
• Alternatives
• Advertising capability through Opaque LSAs or
  using a bit in the (IMMS) byte in DBD packets
  (the first requires an additional mechanism, the
  second is equal to an Options bit)
• Hacking of hello packets for a bit instead of
  employing the RR bit (still requires capability
  announcement)
• Use of timers instead of using the RR bit (can
  be ugly)
• All the above are not among the elegant ways of
  solving this problem

9
To Do

• Finalize the discussion on the list
• Come up with a final version of the draft
  reflecting the consensus
• Update the demand circuit RFC to solve the
  problem this draft addresses
• Move the content on flushing LSAs to a new draft
  and include in this the sending of empty hellos
  as part of the graceful shutdown procedure for
  OSPF routers