Half-Duplex Multicast Distribution Trees (draft-brockners-ldp-half-duplex-mp2mp-00.txt)

1
Half-Duplex Multicast Distribution
Trees(draft-brockners-ldp-half-duplex-mp2mp-00.tx
t)

• IETF 68, March 2007
• Frank Brockners (fbrockne_at_cisco.com)
• IJsbrand Wijnands (ice_at_cisco.com)
• Ali Sajassi (sajassi_at_cisco.com)

2
Problem Statement/Motivation
Client
C1

• Some deployment scenarios (e.g. residential
  wholesale) require transparent and efficient
  multicast transport from a few sources to many
  receivers
• Set of Leaf-LSRs is devided into two
  non-overlapping sets
• Servers
• Clients
• While Servers should be able to communicate to
  all Clients and to each other, Clients should
  not be able to communicate with each other but
  only to Servers

Client
Server
S1
C2
S2
Server
C3
MPLS
Client
Data sent by a Server reaches all other Servers
and Clients.
Client
C1
Client
Server
S1
C2
X
S2
Server
C3
MPLS
Client
Data sent by a Client reaches all Servers but
not the other clients.
3
Solution Objectives

• Simple Configuration and Operation
• Provisioning of a new leaf LSR should not require
  manual configuration changes on any LSR but on
  the leaf LSR
• Automatic setup of the forwarding behavior
  between client and servers (no manual
  configuration requirements on non-leaf LSRs)
• Scalable Implementation
• Limit the amount of state

4
Classic Solution Approach

• Combine P2MP LSPs and MP2P LSPs
• One P2MP LSP per Server (rooted at Server)
• Server to Clients and other Servers communication
• One MP2P LSP per Server
• Client to Server communication
• Each client connects to N x P2MP LSPs and N x
  MP2P LSPs (with N being the number of Servers)
• Requires Send/Receive Operations across multiple
  LSPs for a single Service Instance (due to
  combination of multiple LSPs)
• Addition of a Server requires configuration
  changes on all other Clients and Servers

5
Half-Duplex Multipoint Distribution Trees

• Leverage draft-ietf-mpls-ldp-p2mp-02.txt
• Define extensions to draft-ietf-mpls-ldp-p2mp-02.t
  xt while keeping the spirit (i.e. nomenclature,
  basic principles)
• HD-MP2MP as a new deployment example
• Combine Two LSPs (i.e. Combine two Trees)
• Client Tree All Clients to Servers
• Server Tree All Servers to Clients and Servers
• Leverage P2MP FEC 4 new FEC types
• Similar to MP2MP but this time we need to be
  able to differentiate the origin of the traffic
  (Server or Client)
• Client-Upstream, Client Downstream
• Server-Upsteam, Server Downstream

6
Tree Construction Overview

• Tree construction principles similar to
  draft-ietf-mpls-ldp-p2mp-02.txt for MP2MP
• HD-MP2MP is build as a collection of label states
• Receiver Driven towards Root LSR
• Trees are constructed concurrently (upstream
  label map triggers downstream label map)
• Transit and Root LSRs merge upstream and
  downstream state to optimized forwarding as well
  as half-duplex behavior
• Opaque Value only interpreted by leaf LSRs

S3
C2
R
R
U
Z
S1
C1
S2
7
Next Steps

• Authors appreciate feedback from the WG
• Add example to the draft to help readability