Framework for a Global Hierarchical Multicast

Robert Voigt

Table of Contents

• Stan Barber's Notes
• February 1996 NANOG Home Page

Stan Barber's Notes

Multicast Data Distribution

• Shortest path trees and Center-based trees
  • Shortest perform better
  • Center-based has lower overhead

Some of the solutions that have been proposed:

• MOSPF,DVMRP
• Because of broadcast, then are not really suitable for global use.
• PIM and CBT don't do that.

Scope Control

• Scoping is presently done with ttl. The growth of the Internet
  • is making ttl's mostly useless.
  • here is a need for scoped groups.

Center Location

• A Solution is required.
  • Studies show intelligent location beneficial.
• Present Solutions are currently ad hoc.
  • initial sender
  • administratively selected

Addressing Issues

• Flat Address Space -- No way to filter
  • Proposed use of address ranges
• Name/Address Resolution
  • There is no "DNS" for multicast --?
  • Currently, SD is used.
• Aggregation
  • Multicast addresses inherently aggregated

Interoperability

• Complex issue for present protocols
• Need well defined boundaries for a heterogeneous environment

A Proposal -- A Cluster-based Hierarchical Architecture for Multicast (CHARM)

• Put clusters together concentrically.

Assumptions

• Underlying Unicast Routing based on domains and border routers
• A Level Registrar
  • a hierarchical version of sd
• Multicast border routers

Logical Hierarchy

• The physical paths don't have to follow the hierarchy.

Level "N"

• This means that it is some group (which is may be contained in other groups).

Registrar

• Provides information about MBRs to group members

Control Tree

• Built as a shared tree between MBRs to share cluster related information
  • Similiar to ABR group

Inter-cluster Lists

• Shortest path trees constructed between MBRs at the level of the group
• Designated MBRs are chosen for a cluster on a per source cluster basis

Group Creation

• Group Registration -- via the registrar
• Source Based Trees -- All Local Receivers and partipants in MBRs of this cluster

Joining the group

• Receiver gets list of MBRs from registrar
• Receivers join towards the closest MBR
• MBRs with receivers joined to them join the DMBRs

The Total Data Delivery Path

• Worst case penalties
  • receiving cluster worst case penalty
    • the diameter of the receiving cluster
  • sending cluster worst case penalty
    • the diameter of the receiving cluster
    • and the diameter of the sending cluster

A Simulation was done using a 128 node network with 16 senders and 16 receivers.

Number of hops were counted, but there were no other costs considered.

There appears to be no relationship between the diameter of the clusters versus the diameter of the entire network.

As the network becomes more clustered, the choice of the "center" is less critical and the path length penalty is reduced.

Primary Benefit of this approach

• Scope Control
• Center Location
• Aggregation at the Group Level
• Interoperability (because of well defined boundaries)

Conclusion

• All benefits of the hierarchy at low cost.
• Further improvements can be made.
  • reducing the sending cluster penalty
    • use measure of goodness
  • Reducing the receiving cluster penalty
    • if a congruent unicast hierarchy is present

Questions