Multicast Routing •
Unicast: one source to one destination •
Multicast: one source to many destinations 15-744: Computer Networking •
Two main functions: •
Efficient data distribution •
Logical naming of a group

L-20 Multicast

2

Example Applications Overview •
Broadcast audio/video •
IP Multicast Service Basics •
Push-based systems •
Software distribution •
Multicast Routing Basics •
Web-cache updates •
Teleconferencing (audio, video, shared •
Overlay Multicast whiteboard, text editor) •
Multi-player games •
Reliability •
Server/service location •
Other distributed applications •
Congestion Control

3 4

1 IP Multicast Architecture Multicast – Efficient Data Distribution

Src Src Service model Hosts

Host-to-router protocol (IGMP) Routers

Multicast routing protocols (various)

5 6

Multicast Router Responsibilities IP Multicast Service Model (rfc1112) •
Learn of the existence of multicast groups •
Each group identified by a single IP address (through advertisement) •
Groups may be of any size •
Identify links with group members •
Members of groups may be located anywhere in the Internet •
Establish state to route packets •
Members of groups can join and leave at will •
Replicate packets on appropriate interfaces •
Senders need not be members •
Routing entry: •
Group membership not known explicitly •
Analogy: Src, incoming interface List of outgoing interfaces •
Each multicast address is like a radio frequency, on which anyone can transmit, and to which anyone can tune-in.

7 8

2 IP Multicast Addresses Multicast Scope Control – Small TTLs •
Class D IP addresses •
TTL expanding-ring search to reach or find •
224.0.0.0 – 239.255.255.255 a nearby subset of a group 1 1 1 0 Group ID

•
How to allocated these addresses? s •
Well-known multicast addresses, assigned by 1 IANA 2 •
Transient multicast addresses, assigned and reclaimed dynamically, e.g., by “sdr” program 3

9 11

Multicast Scope Control – Large TTLs Overview

•
Administrative TTL Boundaries to keep multicast •
IP Multicast Service Basics traffic within an administrative domain, e.g., for privacy or resource reasons •
Multicast Routing Basics

The rest of the Internet •
Overlay Multicast

•
Reliability TTL threshold set on interfaces to these links, greater than the diameter An administrative domain of the admin. domain •
Congestion Control

12 13

3 IP Multicast Architecture Multicast Routing •
Basic objective – build distribution tree for Service model Hosts multicast packets •
Multicast service model makes it hard Host-to-router protocol (IGMP) •
Anonymity Routers •
Dynamic join/leave

Multicast routing protocols (various)

14 15

Shared vs. Source-based Trees Source-based Trees

•
Source-based trees Router •
Separate shortest path tree for each sender S Source •
DVMRP, MOSPF, PIM-DM, PIM-SM R Receiver R •
Shared trees R •
Single tree shared by all members R •
Data flows on same tree regardless of sender S

•
CBT, PIM-SM S

R

16 17

4 Shared Tree Shared vs. Source-Based Trees

•
Source-based trees Router • Shortest path trees – low delay, better load distribution S Source
R Receiver •
More state at routers (per-source state) R •
Efficient for in dense-area multicast R •
Shared trees •
Higher delay (bounded by factor of 2), traffic RP R concentration S •
Choice of core affects efficiency S •
Per-group state at routers •
Efficient for sparse-area multicast R •
Which is better?  extra state in routers is bad!

18 19

Routing Techniques Routing Techniques •
Flood and prune •
Core based protocols •
Begin by flooding traffic to entire network •
Specify “meeting place” aka core •
Prune branches with no receivers • Sources send initial packets to core •
Examples: DVMRP, PIM-DM
•
Unwanted state where there are no receivers •
Receivers join group at core •
Link-state multicast protocols •
Requires mapping between multicast group •
Routers advertise groups for which they have receivers address and “meeting place” to entire network •
Examples: CBT, PIM-SM •
Compute trees on demand •
Example: MOSPF •
Unwanted state where there are no senders

20 21

5 Distance-Vector Multicast Routing Example Topology •
DVMRP consists of two major components: G G •
A conventional distance-vector routing protocol (like RIP) •
A protocol for determining how to forward multicast packets, based on the routing table •
DVMRP router forwards a packet if •
The packet arrived from the link used to reach S the source of the packet (reverse path forwarding check – RPF) •
If downstream links have not pruned the tree

G

22 23

Broadcast with Truncation Prune

G G G G

Prune (s,g)

S S Prune (s,g)

G G

24 25

6 Graft Steady State

G G G G G G Report (g)

Graft (s,g)

S Graft (s,g) S

G G

26 27

Overview Supporting Multicast on the Internet

•
IP Multicast Service Basics

•
Multicast Routing Basics Application ? At which layer should multicast be •
Overlay Multicast ? IP implemented? •
Reliability Network •
Congestion Control Internet architecture

28 29

7 IP Multicast End System Multicast

MIT MIT1 Berkeley MIT Berkeley UCSD MIT2 UCSD

CMU1 CMU routers CMU end systems multicast flow CMU2 Berkeley MIT1 Overlay Tree MIT2 •
Highly efficient UCSD •
Good delay CMU1 CMU2

30 31

Potential Benefits Over IP Multicast Concerns with End System Multicast •
Quick deployment •
Self-organize recipients into multicast delivery overlay tree •
All multicast state in end systems •
Must be closely matched to real network topology to be •
Computation at forwarding points simplifies efficient •
Performance concerns compared to IP Multicast support for higher level functionality •
Increase in delay MIT1 MIT •
Bandwidth waste (packet duplication) Berkeley Berkeley MIT1 Berkeley MIT1 MIT2 UCSD UCSD MIT2 UCSD MIT2 CMU1 CMU1 CMU1

CMU IP Multicast End System Multicast CMU2 CMU2

CMU2 32 33

8 Overview Implosion

•
IP Multicast Service Basics Packet 1 is lost All 4 receivers request a resend

S Resend request S •
Multicast Routing Basics 1 2

•
Overlay Multicast R1 R1

•
Reliability R2 R2

•
Congestion Control R3 R4 R3 R4

34 35

Retransmission Exposure

Packet 1 does not reach R1; Packet 1 resent to all 4 receivers •
Re-transmitter Receiver 1 requests a resend

• Options: sender, other receivers Resend request
S Resent packet S •
How to retransmit 1 2 1 •
Unicast, multicast, scoped multicast, 1 retransmission group, … R1 R1 •
Problem: Exposure R2 R2

R3 R4 R3 R4

36 37

9 Ideal Recovery Model Scalable Reliable Multicast (SRM) Packet 1 reaches R1 but is lost Only one receiver sends NACK to before reaching other Receivers the nearest S or R with packet •
Originally designed for wb S Resend request S Repair sent •
Receiver-reliable Resent packet only to 1 2 1 those that •
NACK-based need packet •
Every member may multicast NACK or R1 R1 retransmission 1 R2 R2

R3 R4 R3 R4

38 39

SRM Request Suppression Deterministic Suppression Packet 1 is lost; R1 requests Packet 1 is resent; R2 and R3 no resend to Source and Receivers longer have to request a resend 3d Time Resend request S Resent packet S d data 2d X1 2 1 d d R1 R1

d nack repair = Sender 3d

R2 R2 = Repairer X d 4d = Requestor d

Delay varies Delay = C1
dS,R R3 R3 by distance X 40 41

10 SRM Star Topology SRM: Stochastic Suppression Packet 1 is lost; All Receivers Packet 1 is resent to all Receivers request resends 0 Time Resend request S Resent packet S d data 1 2 1 1 X d session msg repair

d 2 NACK d 3 2d R2 R3 R4 R2 R3 R4 = Sender Delay = U[0,D ]
d 2 S,R = Repairer

Delay is same length = Requestor

42 43

SRM (Summary) Overview •
NACK/Retransmission suppression •
IP Multicast Service Basics •
Delay before sending •
Delay based on RTT estimation •
Multicast Routing Basics •
Deterministic + Stochastic components •
Periodic session messages •
Overlay Multicast •
Full reliability •
Estimation of distance matrix among members •
Reliability

•
Congestion Control

44 45

11 Multicast Congestion Control Video Adaptation: RLM •
What if receivers •
Receiver-driven Layered Multicast have very •
Layered video encoding different 100Mb/s •
Each layer uses its own mcast group bandwidths? 100Mb/s R S •
On spare capacity, receivers add a layer •
Send at max? R 1Mb/s •
On congestion, receivers drop a layer •
Send at min? ???Mb/s 1Mb/s R •
Join experiments used for shared learning •
Send at avg?

56Kb/s R

46 47

Layered Media Streams Drop Policies for Layered Multicast •
Priority R1 joins layer 1, •
Packets for low bandwidth layers are kept, drop R2 joins layer 2 queued packets for higher layers R1 joins layer 3 10Mbps •
Requires router support 10Mbps R2 join layer 1, •
Uniform (e.g., drop tail, RED) S R 512Kbps join layer 2 10Mbps R •
Packets arriving at congested router are fails at layer 3 dropped regardless of their layer 128Kbps •
Which is better? R3 R3 joins layer 1, fails at layer 2

48 49

12 RLM Intuition RLM Intuition

•
Uniform •
Better incentives to well-behaved users •
If oversend, performance rapidly degrades •
Clearer congestion signal •
Allows shared learning •
Priority •
Can waste upstream resources •
Hard to deploy •
RLM approaches optimal operating point •
Uniform is already deployed •
Assume no special router support

50 51

RLM Join Experiment Join Experiments

•
Receivers periodically try subscribing to higher layer Layer •
If enough capacity, no congestion, no drops 4  Keep layer (& try next layer) •
If not enough capacity, congestion, drops 3  Drop layer (& increase time to next retry) 2 •
What about impact on other receivers? 1 Time

52 53

13 RLM Scalability? •
What happens with more receivers? •
Increased frequency of experiments? •
More likely to conflict (false signals) •
Network spends more time congested •
Reduce # of experiments per host? •
Takes longer to converge •
Receivers coordinate to improve behavior

54

14