By Dr Kim Chin Australian Catholic University, North Sydney, New South Wales, Australia

1

By Dr Kim ChinBy Dr Kim ChinAustralian Catholic University, Australian Catholic University,

North Sydney,North Sydney,New South Wales, New South Wales,

AustraliaAustralia

Multicast NetworkingMulticast Networking

2

OutlineOutline What is Multicast?

– Unicast vs Broadcast vs Multicast Why Multicast? IP Multicast Model MBone and its growth Multicast Trees

– Some Logical Multicast Trees generated when the MBone experienced an explosive growth

3

Outline (continue…)Outline (continue…)

IP Multicast deployment status Factors that hinder deployment Beyond IP multicast

– Application layer multicast– Overlay multicast

Summary and the future of Multicast

4

What is What is multicastmulticast?? a receiver-based concept is the sending of messages from one-to-many

receivers or many-to-many receivers is a subset of broadcast

5

Broadcast vs UnicastBroadcast vs Unicast

Broadcast– sends data to everyone whether they want them

or not Unicast

– the common way of transmitting data across the Internet

– sending a single copy of the message to anyone who requests it

6

Network

UC BerkeleyMultimedia Seminar

anhur(Sweden)

ursa(Germany)

cedar (Texas)

alps (Georgia)

tove (Maryland)

law(California)

float(Virginia)

erlang (Massachusetts)

edgar(Washington)Network

Network node

Data packets replicated by this node

Audio packet

7

Problem with Unicast The Internet has traditionally been built to send

information to one person (or computer) at a time--unicast

consider sending information simultaneously to more than one person– e.g. e-mail message to which you've attached a graphics

file unicast--to one person multicast--to many at the same time

8

Why Multicast?Why Multicast? “The unicast delivery paradigm predominant

in today’s Internet does not scale to support the widespread use of video.” [Edwards et. al.]

“Multicast is becoming important because it enables desired applications to scale, ... Widespread use of these applications would be virtually impossible without the scaling provided by multicast services” [Miller]– “push” services of PointCast -- traffic clogged up

networks

9

Why Multicast? (continue 1…)Why Multicast? (continue 1…) Large-scale events:

– (1998), latest version of Microsoft explorer--meltdown

– landing of Pathfinder on Mars--”surfers” overwhelm NASA’s Web site

10

Why Multicast? (continue 2…)Why Multicast? (continue 2…)

– September 11 tragedy, at Northern University, CNN was rebroadcast using multicast on the Internet--over 2000 viewers

– “users on multicast-enabled networks were able to watch real-time video accounts throughout the entire day.” [Edwards et al.]

11

Video Transmission in Unicast & Multicast Video Transmission in Unicast & Multicast Networks Networks (from http://www.mcclellanconsulting.com/)(from http://www.mcclellanconsulting.com/)

12

Multicast Applications

Multimedia

Video Video conferencing Internet audio Multimedia events

Replication: Video and Web servers Kiosks

Content delivery Intranet and Internet

Data-only

Stock quotes News feeds Whiteboarding Interactive gaming

Data delivery Server-server Server-desktop

Database replication Software distribution

13

The IP Multicast ModelThe IP Multicast ModelDeering’s standard multicast model for IP

networks is as follows: IP-style semantics: source sends UDP/IP packets

– no need to register– no schedule transmission– send at any time– best effort

14

The IP Multicast Model The IP Multicast Model (continue...)(continue...)

Open groups:– multicast address known by sender only– need not know group membership– need not be a member of the multicast group they

send packets to– group can have any number of sources

15

The IP Multicast Model The IP Multicast Model (continue...)(continue...)

Dynamic group:– members can join and leave a group– no need to register– no need to synchronize– no need to negotiate with any centralized group

management entity

16

17

18

The IP Multicast backbone The IP Multicast backbone (MBone)(MBone)

The “virtual network backbone” which joins together the multicast-capable portion of the Internet was the MBone.– Originated from DARTNet (DARPA Research

Testbed network) the success of the weekly DARTNet meetings

resulted in extending the multicast infrastructure:

19

MBone (continue…)MBone (continue…)– multicast routing function provided by mrouted

a daemon process--received encapsulated multicast packets-- incoming interface outgoing interface

– tunnels (multicast routers at the edge of a multicast-capable portion of the Internet)

each tunnel connected two end-points via one logical link--crossed several Internet routers

– multicast packet received at a tunnel endpoint--broadcast on a local network

– DVMRP (Distance Vector Routing Protocol)

20

21

22

Multicast TreesMulticast Trees

multicast routing was a controlled form of flooding

no pruning initially -- pruning was deployed several years later

broadcast & prune used to create multicast trees

reverse shortest path tree --rooted the source

23

Multicast Trees (continue 1…)Multicast Trees (continue 1…)

4 steps to creating a multicast tree: source broadcasts packets on its local network

– an attached router receives the packets– sends them through all outgoing interfaces

router receiving a packet performs a RPF (Reverse Path Forwarding) check – router checks incoming interface (packet received)

is used as the outgoing interface to reach the source

24

Multicast Trees (continue 2…)Multicast Trees (continue 2…) When a packet reaches a router (leaf router) with

attached hosts– leaf router checks if there are known group

members on its attached subnets– IGMP queries issued periodically to discover group

members– leaf router:

does nothing if there are group members no group members--leaf router sends a prune towards the

source on the RPF interface

25

Multicast Trees (continue 3…)Multicast Trees (continue 3…)

Prune packets sent back to the source– routers create prune state for the interface

receiving the prune

26

anhur(Sweden)

law(California)

collage(California)

cedar(Texas)

World Radio Network(Washington D.C) zen

(Missouri)

willow(Arizona)

tove(Maryland)

erlang(Massachusetts)

alps(Georgia)

float(Virginia)

ursa(Germany)

Logical Multicast Tree (WRN July 13th, 1995). It is adapted from Yajnik et al. The bold lines provide connections between the “backbone” routers. The thin lines are branches of the tree on

the edge of the network. • is the backbone router and • is the local LAN router.

27

28

World Radio Network(Washington DC)

tove(Maryland)

alps(Georgia)

anhur(Sweden)

pax(France)

ocarina(Kentucky)

cedar(Texas)

erlang(Massachusetts)

Logical Multicast Tree with WRN (December 18th, 1995). • are the backbone routers and • are local LAN routers. The bold lines are the routes which connect some selected intermediate MBone routers between hosts and the WRN root.

29

30

Logical Multicast Tree (April 19Logical Multicast Tree (April 19thth, 1996). It is adapted , 1996). It is adapted from Yajnik et al.from Yajnik et al.

Radio Free Vat (California)

spiff(Sweden)

ursa(Germany)

float(Virginia)

cedar(Texas)

erlang(Massachusetts)

pax(France)

edgar(Washington)

excalibur(California)

bagpipe(Kentucky)

tove(Maryland)

artemis(France)

31

32

law(California)

tove(Maryland)

excalibur(California)

ganef (California)

edgar(California)

float(Virginia)

lupus(Germany)

spiff(Sweden)

cedar(Texas)

Radio Free Vat (California)

erlang (Massachusetts)

Logical Multicast Tree (May 8th, 1996). It is adapted from Yajnik et al. The bold lines represent the connections between backbone routers. All other lines are branches of the tree and they are on the edge of the network, leading to the receiving hosts. is the backbone router and is the local LAN router.

33

Deployment StatusDeployment Status Deployment has been very slow even though

multicast is an old concept by Internet standards—compare with WWW & HTTP

Jan 1992—the MBone, did not exist—in 1995, it made up 20% of all the Internet data bytes at one research lab—40% at another and more than 50% at yet another---but traffic overshadowed by Web soon after.

34

Deployment Status (continue…)Deployment Status (continue…) the Web was quite unknown until late 1992 --

then a “stunning pattern of growth set in”: “a research site’s Web traffic began to double every 6 weeks, and continued to do so for 2 full years!!!” by 1994, Web traffic wholly dominated the site’s activities [Paxson,1996].

Web traffic has overshadowed MBone traffic ever since

35

Factors that hinder deploymentFactors that hinder deployment multicast lacks the “killer” applications

– most popular multimedia applications access the content via multicast then falling over to unicast

lack of scalable inter-domain routing protocol state scalability issue with a large number of groups how to charge for Multicast services—lack of

appropriate pricing model loss of revenues from unicast bandwidth issue of end-to-end connectivity

36

Factors that hinder deployment Factors that hinder deployment (continue…)(continue…)

can the Internet ever have prime-time television quality video?

lack of test tools for trouble-shooting Denial of Service attacks—ASM (Any-Source-

Multicast) lack of security support multicast protocols are complex and may break

the unicast network

37

Beyond IP Layer MulticastBeyond IP Layer Multicast

Application layer multicast group membership, tree construction, data

forwarding controlled by end hosts, thus requiring no support from intermediate nodes such routers

Overlay multicast– Multicast functionalities supported by some

additionally deployed intermediate nodes forming an overlay network

38Taken from [Lao et al, 2005]

39

Application Layer MulticastApplication Layer Multicast

ALMI (Application Level Multicast Infrastructure) [Pendarakis et al]– provides a multicast middleware which is

implemented above the socket layer– scales for a large number of groups with

number of members small– independent of multicast support in routers

40

ALMIALMI

An ALMI session consists of a session controller and multiple members– a session controller is in a location where it is

easily accessible by members– session members are organized into a multicast

tree– unicast connection between 2 members is

represented by a link

41

Application Layer Multicast Application Layer Multicast ApproachApproach

Multicast related features are implemented at end hosts

Data packets are transmitted between end-hosts via unicast and replicated at end hosts

Can be deployed with ease because it does not require infrastructure support from intermediate nodes [Lao et al]

42

Disadvantages of Application-layer Disadvantages of Application-layer MulticastMulticast

Not scalable to support large multicast groups due to its rather low bandwidth efficiency

Tree maintenance at end hosts causes heavy control overhead

Difficult for ISPs to have an effective profit-making service model because group membership and multicast trees managed at end hosts—hard to have member access control as well as knowledge of a group’s bandwidth usage [Lao et al]

43

Overlay MulticastOverlay Multicast Two-tier Overlay Multicast (TOM) [Lao et al]

– A key feature is its backbone service domain, MSON (Multicast Service Overlay Network), consisting of service nodes or proxies strategically deployed by MSON provider (ISP)

– The design of MSON relies on well-defined relationship between:

the MSON provider; the network service provider; and the group coordinators

44

SummarySummary

Multicast is a more efficient mode of transmitting packets, especially video packets, than unicast and broadcast– Microsoft supports Multicast project in China

45

Summary (continue 1…)Summary (continue 1…)

– “Microsoft to “fully support” and “co-operate” with IP-Set-Top-Box manufacturer World Multicast China, on first ever direct to home IP Multicast service. This test to take place in the city of Shaoxing China beginning in early 2005.” [http://www.ipmulticast.com/]

46

Summary (continue 2…)Summary (continue 2…)

– “Any time you are dealing with an emerging technology that has infinite potential you are going to have interested parties, our secure reliable IP multicast technology is the toolbox of missing links that have thus far hindered Inter-domain IP Multicast from proliferation. Now this test will change all of that. “

47

Summary (continue 3…)Summary (continue 3…)

– “IP Multicast is the only technology built-into the current inter-domain routing infrastructure that addresses mass media on the Internet.We have dubbed our technology “Infinicast” because of its ability to support an infinite number of users from just one stream of video or audio.”[World Multicast Technology inventor Ian A. Stewart ]

48

The file size required to represent about one minute of media (taken from ASTD’s source for E-learning)

49

Summary (continue 4…)

IP Multicast Model IP Multicast Routing Protocols The MBone and its growth Multicast trees IP multicast deployment status Factors hindering IP multicast deployment

50

Summary (continue 5…)

Beyond IP multicast– Application layer multicast– Overlay multicast

51

ReferencesReferences B. M. Edwards, L. A. Giuliano and B. R. Wright. Interdomain Multicast Routing: Practical

Juniper Networks and Cisco Systems Solutions. Addison Wesley Professional, April 2002. C. K. Miller. Multicast Networking and Applications. Addison-Wesley Longman, Inc.,

Massachussetts, January 1999. K.C. Almeroth. The Evolution of Multicast. (www.stardust.com) Developing Media for Low Bandwidth. (ASTD’s Source for E-Learning) V. Paxson. Why We Don’t Know How To Simulate The Internet. In Proceedings of the 1997

Winter Simulation Conference, Atlanta, GA, 1997. M. Yajnik, J. Kurose and D. Towsley. Packet Loss Correlation in the Mbone Multicast

Network. UMASS CMPSCI Technical Report # 96-32. http://www.ipmulticast.com/ L. Lao et al. A Comparative Study of Multicast Protocols: Top, Bottom, or in the Middle?

Technical Report TR040054 (2005) Computer Science Department UCLA. L. Lao et al. A Scalable Overlay Multicast Architecture for Large-Scale Applications.

Technical Report TR040008 (2004) Computer Science Department UCLA. D. Pendarakis et al. ALMI: An Application Level Multicast Infrastructure. WUCS-11-17

Department of Computer Science, Washington University.