Browse Prior Art Database

MULTICAST PACKET DISTRIBUTION VIA DROP-AND-FORWARD

IP.com Disclosure Number: IPCOM000235739D
Publication Date: 2014-Mar-24
Document File: 6 page(s) / 43K

Publishing Venue

The IP.com Prior Art Database

Related People

Stewart Bryant: AUTHOR [+3]

Abstract

Techniques are presented herein for multicasting a packet using a drop-forward technique in situations where a packet is forwarded based on an explicit path encoded in its header. This enables moving the multicast state from the network to the ingress router and the packet itself.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 38% of the total text.

Page 01 of 6

MULTICAST PACKET DISTRIBUTION VIA DROP-AND-FORWARD

AUTHORS:

Stewart Bryant

Dan Frost

Giles Heron

CISCO SYSTEMS, INC.

ABSTRACT

    Techniques are presented herein for multicasting a packet using a drop-forward technique in situations where a packet is forwarded based on an explicit path encoded in its header. This enables moving the multicast state from the network to the ingress router and the packet itself.

DETAILED DESCRIPTION

    In networking systems, it is important to reduce the state held in the network core for supporting multicast-like services. Multicast requires that each hop (node) be pre- programmed with a method of identifying each multicast packet that passes through the node, together with the set of next hops to which it is to be forwarded. This state is expensive both in terms of the state itself (a set of next hops for every multicast group that is currently active) and in terms of the protocol overhead required to create, maintain, and manage the state.

    Vector switching and related packet encoding schemes such as segment routing minimizes core network state by placing the path state in the packet itself. This works well for unicast traffic because a path is described in terms of the set of hops that the packet needs to traverse to reach its destination. The identifier for each next hop is well- known at the hop that needs to forward the packet. On the other hand, to do multicast the domain name for the whole multicast tree needs to be encoded the entire tree in the packet header needs to be encoded as a tree structure. The latter however does not have a

Copyright 2014 Cisco Systems, Inc.

1


Page 02 of 6

mapping to well-known forwarding structures (such as Multi-protocol Label Switching or Internet Protocol), and scales as the number of branches in the network.

    By some means outside the scope of this paper, the source of the packet discovers the set of links and nodes that comprise the network, and it discovers the set of destinations to which it must forward the packet. The network topology may be discovered by an Internal Gateway Protocol (IGP) or by a Software-Defined Network (SDN) controller. The set of destinations may be discovered by some request, by some advertisement protocol or by an SDN controller, as may be the method of mapping some characteristic of the ingress packet to the set of destinations.

    A hamiltonian path is a path that traverses all nodes in the sub-network exactly once. In the simplest form of this techniques presented herein, the source calculates a hamiltonian path through the set of destinations, and pushes a stack of adjacency labels such that each has the property "forward to a specific next hop" + drop/no-drop. In other words, there exist two sets of adjacency labels: one with the drop property and one with the no-drop property. The packet is then forwarded in sequence to each node of the sub- network via the encoded hamiltonian path. This very simplistic approach does not scale well, but...