Graceful restart why

1
Graceful restart - why

• A graceful restart mechanism for ANCP needs to be
  defined as per ANCP framework and it is also
  referred to in the ANCP draft.
• ANCP runs on TCP and the ANCP session is brought
  down whenever TCP session resets. This wipes out
  all the session data on both AN and NAS.
• The quicker the AN detects the loss of adjacency,
  the faster it can connect with a NAS.
• We present the above problems and their initial
  solutions.

2
What data is transported and built in ANCP
sessions

• Capabilities four categories of capabilities
  are exchanged for the session and are fixed for
  the session
• Topology data NAS discovers the topology of DSL
  ports on AN and builds a topology database
• Multicast flow data This data on NAS consists of
  how many S,G sessions are active per subscriber
  on AN, what is the bandwidth usage of each flow,
  etc

3
What have we identified as issues to address

• Topology Discovery After a fresh adjacency is
  established the topology information that NAS
  builds is not complete
• AN doesnt send port-up data about all active
  ports after a new session is established
• NAS may have lost all of its topology information
  as a result
• Detection of Adjacency Loss The AN needs to wait
  till the ANCP Hellos expire to detect adjacency
  loss
• Multicast Upon a restart of the adjacency the
  multicast flow information on the NAS may not be
  synchronized with that of AN.

4
How to solve them

• Topology Discovery Have the AN send port-up for
  all active ports upon a fresh session
  establishment.
• Needs very little change in the protocol
  operation on AN and doesnt need any new
  messages. NAS can correctly rebuild the topology
  information.
• Detection of Adjacency Loss TCP Socket state can
  be used to detect ANCP transport failure more
  quickly.
• Multicast NAS can request an all flow report
  from AN using existing messages to recover flow
  info.
• Other multicast cases (like which ones need
  re-authentication and which dont need
  re-authentication)TBD in future revisions.

5
Summary

• Solve existing protocol issues in line with
  protocol framework
• Use existing message types with minimal changes
  to protocol mechanism.
• Doesnt rely on TCP HA for correct protocol
  operation doesnt make assumptions about
  whether TCP state-full HA is present or not

6
Proposal

• Soliciting feedback.