Multicast Final

8/2/2019 Multicast Final

1/31

Multicast Routing


2/31

The Basic Idea

IP Multicast traffic for a (source, destinationgroup) pair is transmitted from the source tothe receivers via a spanning tree that

connects all hosts in the group. Differentrouting protocols use different techniques tobuild the trees. Once constructed allmulticast traffic is distributed through the tree


3/31

Communication Tiers

Locally: host informs local multicast router ofdesire to join the group (using IGMP)

Globally: local routers interact with otherrouters to receive multicast packets (usingDVMRP, MOSPF, PIM)


4/31

Introduction: Uses

Bulk data transfer (software updates, mailinglist distribution, stock updates)

Streamed Continuous Media (audio/visualconferences)

Shared Application Data (shared whiteboard)

Interactive Gaming or Simulations (very

intensive)


5/31

The Mbone

Collection of multicast subnetworks(islands)interconnected (tunneled) through theInternet for faster multicast deployment

Started in 1992 with 40 subnets; up to 2800subnets in 1996

Based primarily on UDP for end-to-end

transmission, IGMP for group management,and DVMRP for routing

Use sd (session directory) to get a listing ofactive multicast sessions


6/31

Multicast Propertiess

Must service a variety of applications (somehave specific timing needs)

The overall group should not be effected byindividual join/leave operations

Connectionless

Current Best-effort service


7/31

Two Scenarios

Dense ModeGroup members are densely distributed throughout network(i.e. Many subnets contain at least one group member)

Bandwidth is plentifulUses flooding to propogate information

Protocols include: DVMRP, MOSPF, PIM-DM

Sparse ModeGroup members are sparsely distributed throughout network

Bandwidth is restricted (such as ISDN, home users)

Uses selective techniques to set up and maintain multicasttrees

Protocols include: CBT, PIM-SM


8/31

What defines a group?Network Group: logical grouping to controlmembership in local group (group address,group identifier, group properties, and groupmanagement)

Social Group: participants in thecommunication

Group management procedures advertisegroups to potential members, broadcastrouting information to multicast nodes, andcontrol various group properties


9/31

IGMP

The host sends IGMP report when anapplication joins the multicast group

IP_ADD_MEMBERSHIP socket option

host not required to explicitly "unjoin" group when leaving

The router sends IGMP query at regularintervals which hosts belonging to themulticast group are obligated to reply


10/31

IGMP Evolution

IGMPv1:

Router sends HostMembership Query

msg broadcast onLAN

Host replies toindicate groupmembership

Randomized delaybefore respondingImplicit leave viano reply toQuery(timeout)

RFC 1112

IGMPv2:

Added Group-SpecificQuery

Elected routerqueries to see ifany hosts left ingroup

Leave GroupMessage

Last host replyingto query can sendexplicit LeaveGroup msg

RFC 2236


11/31

IGMP Evolution (cont.)

IGMPv3:

Lets source know if nobody is listening

Receivers can selectively choose sources to listen to(source pruning)Still being designed


12/31

Virtual Circuit Group Mgmt

Group of endpoints named at circuit setuptime

Master may be able to later add newreceivers or senders to multicast circuit

Challenges in reducing state informationrequired when adding new receivers


13/31

Routing Algorithms: The

RequirementsMinimize network load. Need to optimizenetwork resources while avoiding loops andconcentrated traffic

Reliable transmission: data deliveryshouldn't be affected by routing changes, linkfailures, etc.

Use optimal routes as determined byresource availability, bandwidth, nodeconnectivity, price paid

Minimize router state (especially for larger

groups)


14/31

The Building Blocks

Most routing protocols use these basic treealgorithms

Reverse Path Forwarding (one tree per source)

Flood and prune

Steiner Trees (a minimal spanning tree, which isshared for entire group)

Core-based Trees (one tree per group)


15/31

Desirable Characteristics

Flooding can be danagerous inheterogeneous networks. Have toperiodically flood if new receiver appears

Recall the challenges for routing algorithms:Evolve with group membership. Algorithm should uniquelyidentify group members

Minimize state information in the nodes

Optimize routing given cost considerationsAvoid traffic concentration


16/31

Reverse Path Forwarding

Rely on router's existing knowledge ofunicast routing table (shortest path from selfto sender)

Best for densely distributed receivers

Result of algorithm is are directed graphs:source-rooted delivery trees emanating from

the subnet directly connected to sourceEfficient and easy to implement but doesn'tuse group membership information to buildthe tree so packets may hit subnets without

members


17/31

RPF with Pruning

Use RPF technique plus record groupmembership.

Timer controlled; have to periodically floodand re-prune to capture new members

Requires state information per source andper group to be kept at each node (leads to

scalability problems)


18/31

Steiner Trees

Minimum cost tree connecting all routers withattached group members; network resourcesare globally optimized

NP-complete problem; but good heuristicsexist

Forms an undirected tree ==> symmetric

linksGenerally not used in practice:

computational complexity

requires knowledge of entire network topology

must be rerun when a router needs to leave/join


19/31

Core-Based Trees

Single tree shared by everybody(multiple sender / multiple receiver)

One router is labeled as center of tree (but

finding center is NP-complete)Receiver based approach; packets naturallylimited to group members

Best for sparse receiver distributionState only kept per group (not forsource/group pairs)

Can lead to traffic congestion


20/31

Algorithms Summary

Most algorithms include challengingparameters (network topology, groupdynamics, member location, other routing

algorithms)Requirements for the optimal algorithm:

Should be transparent to the members that remain in agroup

Should maintain properties of the original route

Should not perturb ongoing data transfers

Should be receiver driven (so it adapts well tochanging group membership)


21/31

Routing Protocols

Distance-vector multicast routing protocol(DVMRP): flood and prune

Multicast Open Shortest Path First (MOSPF):source-based trees (RPF)

Protocol-Independent Multicast - DenseMode (PIM-DM): flood and prune

Protocol-Independent Multicast - SparseMode (PIM-SM): core-based trees

Inter-domain Multicast Routing (IDMR):border gateway multicast protocol


22/31

DVMRP

RFC 1075 (original RFC)

An extension to RIP: multicast routers share

reverse path distances to build the (source-based) delivery tree for each group then usepruning

Use TTL to define distance packet will travel

Hierarchical DVMRP in the works: defineregions within which any protocol can beused and DVMRP is used between regions(not yet deployed)


23/31

MOSPF

RFC 1584

Designed for a single routing domain;

depends on OSPF for unicast routingRouters build efficient shortest path source-based trees without initially flooding

Higher link effieiency (routers can discard

packets immediately based on TTL valuesComputation on demand (only when the firstpacket from source to group is received)


24/31

PIM-DM

Used when group membership is plentifulwithin a region of an internet

Like DVMRP but imports unicast routesinstead of calculating for self (thus theindependence)


25/31

PIM-SM

Used when group members are spread thinlyacross regions of an internet

Join messages are sent to rendezvous pointsto meet new sources

Uses shared trees


26/31

Mobile Hosts

When mobile unit wants to send to group, ituses RPF-based approach from the mobilesupport router

When mobile unit wants to receive amulticast packet, the packet is sent to thewired address of the mobile unit thenforwarded to the mobile address

Main Problems: 1. Mobile source packetsmay not reach all group members; 2. Mobilehost may see delays as enters a new cell; 3.TTL can limit cell reachability


27/31

ATM EnvironmentsIntroduced point-to-multipoint in UNI for audioconferences

Multigroup address concept doesn't exist inATM; sender has to know all members of thegroup

Only VC root node may add or remove leafnodes

Various Approaches: MPOA(multi-protocolover ATM), VC mesh (sender to allmembers), MCS per cluster, SMART, MARS,Lan Emulation


28/31

Traffic ControlProblems introduced by multicast:

Retaining capabilities for heterogeneity

Defining fairness and relative fairness between unicastand multicast

Timescales for congestion controlScaling of control traffic and techniques

Design Issues include:Scalable feedback mechanism

Congestion detection by receivers not sendersProbabilistic query/reply schemes, random delayresponses

Number control packets fixed WRT data packets

RTP, RLM (receiver-driven layeredmulticast)


29/31

End to End Concerns

Reliability: don't want every receiver to ACKthe sender; better solution is to use NACK

Reliability protocols include: Virtual tokenring-based protocol, XTP, SRM

QoS options for group communication:Parameters defined by sender with no negociation

Parameters equal to minimum of each group memberSenders send with higher QoS and each receiver controls forself

Resource reservation very complex

QoS overall not an easy problem


30/31

Scalability

Need to make join and leave operations asefficient as possible

Hierarchical group addressing (and routing)schemes are being developed


31/31

security

Efficient key distribution scheme for multiplerecipients

Group communication leads to greateropportunity for traffic analyzers, denial ofservice attacks, covert channel signaling

May become too complex to understand link

usage (for filtering rules)

Multicast Final

Documents