IP Multicast September 20, 2001 Multicast 2 Overview - - PowerPoint PPT Presentation

▶
ip multicast
SMART_READER_LITE
LIVE PREVIEW

IP Multicast September 20, 2001 Multicast 2 Overview - - PowerPoint PPT Presentation

Jan 10, 2024 201 likes •737 views

Multicast 1 IP Multicast September 20, 2001 Multicast 2 Overview applications models host APIs LAN (IGMP, LAN switches) intra-domain routing inter-domain routing address allocation Additional references (some are

slide-1
SLIDE 1

Multicast 1

IP Multicast

September 20, 2001

slide-2
SLIDE 2

Multicast 2

Overview

➤ applications ➤ models ➤ host APIs ➤ LAN (IGMP, LAN switches) ➤ intra-domain routing ➤ inter-domain routing ➤ address allocation Additional references (some are dated!):

  • Stephen A. Thomas, IPng and the TCP/IP protocols, Wiley, 1996.

September 20, 2001

slide-3
SLIDE 3

Multicast 3

  • Christian Huitema, Routing in the Internet, Prentice Hall, 1995.
  • Crowcroft/Handley/Wakeman, Internetworking Multimedia, 2000.

Partially drawn from http://www-scf.usc.edu/˜dbyrne/960223.txt (D. Estrin) September 20, 2001

slide-4
SLIDE 4

Multicast 4

Broadcast and multicast

broadcast: all hosts on (small, local) network directed broadcast: all hosts on remote network multicast: multiple recipients (group)

September 20, 2001

slide-5
SLIDE 5

Multicast 5

Applications for Multicast

  • audio-video distribution (1-to-many) and symmetric (all-to-all)
  • distributed simulation (war gaming, multi-player Doom, . . . )
  • resource discovery (where’s the next time server?)
  • file distribution (stock market quotes, new software, . . . )
  • network news (Usenet)

September 20, 2001

slide-6
SLIDE 6

Multicast 6

Multicast trees

spanning tree ≡ tree that connects all the vertices (hosts/routers) shared tree: single tree for all sources S

  • minimum-cost spanning (MST) tree (where cost = hops, delay, $, . . . )
  • does not minimize length of S to individual destination
  • all traffic concentrated on tree ➠ reservation failures

per-source tree: build independently for each source ➠ many variations!

  • September 20, 2001
slide-7
SLIDE 7

Multicast 7

Steiner Tree

Minimizes the total number of links for all sinks

  • N-P complete (travelling salesman), unstable: small additions → large changes in

traffic flows

  • Add one node:
  • September 20, 2001
slide-8
SLIDE 8

Multicast 8

Finding MST via Prim’s Algorithm

  • centralized, finds MST for G = (V, E)
  • U: set of vertices connected, start with one
  • add lowest-cost edge (u, v) with u ∈ U and v in V − U.
  • T ← T ∪ (u, v)
  • U ← U ∪ v

September 20, 2001

slide-9
SLIDE 9

Multicast 9

Connection-oriented multicast

  • enumerate sources explicitly ➠ source-based trees
  • examples:

– ATM ➠ explicitly add each end point – ST-II ➠ enumerate end points in setup message – ATM, ST-II: end nodes attach themselves to tree – enumeration of end points in packet

  • only connection-oriented (packet header size!)
  • source needs to know destinations ↔ resource discovery, dynamic groups difficult
  • but: natural transition from unicast to multicast

September 20, 2001

slide-10
SLIDE 10

Multicast 10

ST-II

  • IEN 199: ST ➠ ST-II: RFC 1190 (1990) ➠ ST-II+: RFC 1819 (1995)
  • hard state
  • combines building tree with resource reservation
  • first Internet resource allocation protocol
  • sender-initiated tree ➠ receiver-initiated joins ST2+

September 20, 2001

slide-11
SLIDE 11

Multicast 11

Host group model

Deering, 1991:

  • senders need not be members;
  • groups may have any number of members;
  • there are no topological restrictions on group membership;
  • membership is dynamic and autonomous;
  • host groups may be transient or permanent.

September 20, 2001

slide-12
SLIDE 12

Multicast 12

Local multicast

Some local networks are by nature multi/broadcast: Ethernet, Token Ring, FDDI, . . . Ethernet, Tokenring:

  • broadcast: all ones
  • multicast: 01.xx.xx.xx.xx.xx
  • adapter hardware can filter dynamic list of addresses

ATM: point-to-point links ➠ need ATM multicast server

September 20, 2001

slide-13
SLIDE 13

Multicast 13

IP multicast

  • host-group model
  • network-level; data packets same, only address changes
  • need help of routers
  • special IP addresses (class D): 224.0.0.0 through 239.255.255.255
  • 28 bits ➠ 268 million groups (plus scope)
  • 224.0.0.x: local network only ➠ 224.0.0.1: all hosts; 224.0.0.2: all routers
  • some pre-assigned (224.0.1.2: SGI Dogfight)
  • others dynamic (224.2.x.x for multimedia conferencing)
  • map into Ethernet: 01.00.5E.00.00.00 + lower 23 bits
  • ttl value limits distribution: 0=host, 1=network

September 20, 2001

slide-14
SLIDE 14

Multicast 14

Administrative Scoping

  • address-based
  • 239.255/16: IPv4 local scope
  • 239.192/14: organization local scope
  • relative addresses (from top) for common applications within scope

September 20, 2001

slide-15
SLIDE 15

Multicast 15

Multicast programming

UDP, not TCP (obviously. . . ) struct sockaddr_in name; struct ip_mreq imr; sock = socket(AF_INET, SOCK_DGRAM, 0); imr.imr_multiaddr.s_addr = htonl(groupaddr); imr.imr_interface.s_addr = htonl(INADDR_ANY); setsockopt(sock, IPPROTO_IP, IP_ADD_MEMBERSHIP, &imr, sizeof(struct ip_mreq)); name.sin_addr.s_addr = htonl(groupaddr); name.sin_port = htons(groupport); bind(sock, &name, sizeof(name)); recv(sock, (char *)buf, sizeof(buf), 0);

September 20, 2001

slide-16
SLIDE 16

Multicast 16

IGMP

Multicast for local (broadcast) networks, between router and hosts

R

(Ethernet, FDDI, Tokenring, ...) multicast-capable medium

H H H R

querier Internet non-querier 128.59.27.35 128.59.27.17

  • router listens to all multicast packets on all interfaces
  • hosts sends IGMP report for first process to join group to that multicast group

(ttl=1), maybe repeat

  • router multicasts query to all hosts (224.0.0.2) ≈ every 125 seconds or on start-up
  • host waits and listens for others; if nobody else, send response for groups it’s in

September 20, 2001

slide-17
SLIDE 17

Multicast 17

  • if “responsible” for group, notify “all router” group ➠ querier sends

group-specific query ➠ reduce bandwidth consumption

  • random interval determined by router (< 10 seconds)
  • really appropriate for today’s switched Ethernet?

September 20, 2001

slide-18
SLIDE 18

Multicast 18

IGMPv2 timing

General query (GQ) Membership report (MR) Leave group (LG) Group-specific query (GSQ) all routers all systems 10 sec. querier host 1 sec. group host joins group host leaves group MR GQ GSQ LG MR 10 sec.

September 20, 2001

slide-19
SLIDE 19

Multicast 19

IGMPv2 packet

version

16 32 24 8 4 12

16-bit checksum 8 bytes IGMP 32-bit group address (class D IP address) (2) type (1,6,7) response time

$ netstat -g Group Memberships Interface Group RefCnt

  • -------- -------------------- ------

lo0 ALL-SYSTEMS.MCAST.NET 1 le0 224.2.127.255 1 le0 ALL-SYSTEMS.MCAST.NET 1

September 20, 2001

slide-20
SLIDE 20

Multicast 20

IGMPv3

  • adds source filtering to IGMPv2
  • Membership Report includes lists of sources to include or exclude
  • Group-and-Source-Specific Query asks whether anybody cares about the group

and the sources listed

  • unlike IGMPv2, host no longer suppresses membership reports if it hears from

another host – accounting – avoid Ethernet switches having to remove “outbound” IGMP reports to fool hosts – for efficiency, single membership report can list multiple groups Note: IPv6 defines new protocol, Multicast Listener Discovery (MLD)

September 20, 2001

slide-21
SLIDE 21

Multicast 21

Reverse path flooding

iif: incoming interface; oif: outgoing interface

  • if iif is on shortest path to source S
  • forward to all other oifs (RPF check) towards receivers R in group G
  • avoids forwarding duplicates

September 20, 2001

slide-22
SLIDE 22

Multicast 22

Multicast forwarding

First packet (truncated broadcast)

S R

  • router

host message no member

  • n local net

not shortest path x

September 20, 2001

slide-23
SLIDE 23

Multicast 23

Reverse path broadcasting

  • do RPF check as before
  • exchange unicast routing info to establish “parentage”
  • restrict oifs to child nodes

➠ reduce duplicates

September 20, 2001

slide-24
SLIDE 24

Multicast 24

Multicast routing

  • link-state based
  • dense mode
  • sparse mode

September 20, 2001

slide-25
SLIDE 25

Multicast 25

Multicast forwarding with truncation

  • flood with RPF check
  • pruning: leaves of tree send “prune” if no members below
  • receivers tell routers of membership
  • routers know whether to forward to LAN or prune
  • prune state must time out ➠ periodic broadcast
  • trade-off: join latency ↔ bandwidth
  • add: explicit “graft” to cancel prune: ➠ join latency ↓
  • still need occasional broadcast for topology changes

September 20, 2001

slide-26
SLIDE 26

Multicast 26

Multicast forwarding

With pruning:

S R

prune 2nd message

  • ➠ router needs to keep “negative” list for groups

September 20, 2001

slide-27
SLIDE 27

Multicast 27

Distance Vector Multicast Routing Protocol (DVMRP)

  • flood + RPF check
  • pruning: time out 1 minute
  • routers may send grafts upstream
  • only send to children
  • maintain routing information (DV)
  • used in old MBone overlay network

September 20, 2001

slide-28
SLIDE 28

Multicast 28

Multicast Open Shortest Path First (MOSPF)

  • link-state based
  • include membership info in link-state advertisements
  • compute tree for each S, G pair ➠ oifs
  • can create shortest-path trees even with asymmetric links
  • cannot afford to recompute trees with each LS change

September 20, 2001

slide-29
SLIDE 29

Multicast 29

PIM-DM

  • use unicast routing table
  • DVMRP: include only oif that use this router to reach source
  • PIM-DM: forward to all outgoing interfaces

September 20, 2001

slide-30
SLIDE 30

Multicast 30

Problems

  • “multicast storms”
  • MOSPF: broadcast of membership to off-tree areas
  • DVMRP: occasional broadcast of packets ➠ bad for WANs
  • prune state in routers for sparse groups
  • multicast routing vs. unicast routing: reverse path with asymmetric links
  • hierarchical routing?
  • few “big” senders, lots of background mumbling

➠ compromise on optimal trees

September 20, 2001

slide-31
SLIDE 31

Multicast 31

Protocol Independent Multicast (PIM-SM)

  • uses unicast routing
  • supports SPTs and shared trees (rooted at “rendezvous point” RP), depending on

traffic

  • 1. group-specific RP-rooted shared tree
  • 2. source-based tree

September 20, 2001

slide-32
SLIDE 32

Multicast 32

PIM-SM: RP election

  • RP selected by hash of G
  • bootstrap router (BSR) candidate sends list of candidate RPs
  • candidate BSRs, configured with priority
  • multicast candidacy locally (ttl = 1), then flood
  • elected routers periodically sends bootstrap message with RPs
  • {candidate BSR} ≈ {candidate RP}
  • candidate-RP sends message to BSR

September 20, 2001

slide-33
SLIDE 33

Multicast 33

PIM-SM: shared tree

  • send packet via unicast in “register” message, encapsulated, to RP
  • RP forwards message down shared tree
  • receivers send “join” to RP to join shared tree
  • joins stop when reaching tree, install (∗, G) state

September 20, 2001

slide-34
SLIDE 34

Multicast 34

PIM-SM: source-specific tree

  • 1. bypass encapsulation
  • RP sends “join” towards S
  • nodes recognize destination and forward based on G
  • 2. receivers join
  • 3. and prune shared tree for S

September 20, 2001

slide-35
SLIDE 35

Multicast 35

PIM-SM

C prune prune Join A! Join

R3

E D

RP S R1 R2

F A I B H G

September 20, 2001

slide-36
SLIDE 36

Multicast 36

Sparse Mode Problems

  • single point of failure
  • hot spot
  • non-optimal path
  • complexity

September 20, 2001

slide-37
SLIDE 37

Multicast 37

Interdomain sparse multicast routing: CBT

  • core-based trees: bidirectional center-based shared trees routed at core
  • receivers send join messages to core
  • senders send data to core, but can be short-cut −

→ send to all interfaces participating in group

  • no SPTs
  • hard-state with acknowledged join from core or first on-tree router

+: no source specific state −: path lengths, traffic concentration

  • explicit joining (vs. implicit join and explicit prune)

join messages from R’s router to root of tree

  • not much implementation

September 20, 2001

slide-38
SLIDE 38

Multicast 38

MBONE

  • MBONE ≡ multicast backbone
  • overlay network over Internet, up to 10,000 routes
  • difficulty of limiting fan-out
  • needed until deployment of multicast-capable backbone routers
  • IP-in-IP encapsulation ➠ tunneling:

4 (IP) 17 (UDP) 192.1.2.3 128.3.5.6 193.1.1.1 224.2.0.1 UDP RTP audio/video data source: 193.1.1.1; group: 224.2.0.1; MBONE tunnel: 192.1.2.3 to 128.3.5.6 IP header IP header

  • limited capacity, resilience

September 20, 2001

slide-39
SLIDE 39

Multicast 39

Mbone

encapsulated 192.1.2.3 193.1.1.1 non-multicast router multicast-capable router 128.3.5.6 Mbone router (workstation)

September 20, 2001

slide-40
SLIDE 40

Multicast 40

Inter-domain multicast

  • one RP per AS
  • use intra (interior) routing protocol, like PIM-SM or DVMRP
  • two approaches to scaling:

– short term: MSDP = distribution of sender information – longer term: BGMP = shared inter-domain tree

September 20, 2001

slide-41
SLIDE 41

Multicast 41

Multicast Source Discovery Protocol (MSDP)

  • join together PIM-SM regions (“AS”)
  • RPs use MSDP to discover sources in other regions
  • and can send them them PIM “join” requests if there are local receivers
  • thus, each inter-domain source gets source tree

September 20, 2001

slide-42
SLIDE 42

Multicast 42

MSDP Operation

  • MSDP RPs peer with fellow RPs via TCP
  • periodically send “source active” to peer RPs: (source address, group, RP)
  • flood “source active” message in RPF style
  • peers can aggregate messages
  • first few data messages can be exchanged via MSDP peers (encapsulated)
  • works reasonably well only when few senders

September 20, 2001

slide-43
SLIDE 43

Multicast 43

MSDP Operation

sender BGP peer SA messages MSDP peer MSDP peer SA message

(S.224.2.0.1,RP(A))

RP RP RP RP RP RP AS A S RP RP

September 20, 2001

slide-44
SLIDE 44

Multicast 44

Border Gateway Multicast Routing Protocol (BGMP)

  • bidirectional shared tree for each group
  • TCP connections between routers (external BGP peers)
  • root domain
  • distribute “routes” to AS hosting core
  • packets can bypass BGMP core
  • packet forwarding similar to PIM-SM RP

September 20, 2001

slide-45
SLIDE 45

Multicast 45

BGMP

AS1 AS2 BGMP M-IGP AS3 AS4 AS5 BGMP M-IGP BGMP M-IGP B G M P M

  • I

G P BGMP M-IGP B G M P M

  • I

G P BGMP M-IGP BGMP M-IGP B G M P M

  • I

G P BGMP M-IGP BGMP M-IGP BGMP M-IGP

1 2 3 4 5 6 7 8 9 10 11 12

September 20, 2001

slide-46
SLIDE 46

Multicast 46

Multicast address allocation

hierarchical, with different time scales:

  • 1. intra-domain, clients contact local MAAS server in domain via MADCAP
  • 2. MAAS gets it via Multicast Address Allocation Protocol (AAP) from MASC
  • MASC routers multicast availability to the MAAS
  • MAAS multicast claims
  • 3. MASC divide space for inter-AS for large blocks

bypass inter domain: assign 233/8 for per-AS static allocation

September 20, 2001

slide-47
SLIDE 47

Multicast 47

Multicast Address Allocation

MADCAP AAP MASC

AAP

AS AS AS AS MASC

AAP

AAP

MADCAP

September 20, 2001

slide-48
SLIDE 48

Multicast 48

Multicast Address Dynamic Client Allocation Protocol: MADCAP

RFC 2730

  • UDP-based request-response (similar to DHCP)
  • one or more local servers
  • may request addresses in the future
  • specify maximum delay
  • can request specific address
  • discover scopes via INFORM
  • multicast request via DISCOVER
  • server hands out, client confirms via REQUEST
  • expires or via RELEASE

September 20, 2001

slide-49
SLIDE 49

Multicast 49

MADCAP

DISCOVER ACK ACK REQUEST ACK RELEASE ACK multicast unicast CLIENT MAAS1 MAAS2 September 20, 2001

slide-50
SLIDE 50

Multicast 50

AAP: Multicast Addresses within AS

AAP Multicast Group

Autonomous System

ASA ASA ACLM AIU AITU ASRP MASC MAAS MASC router

September 20, 2001

slide-51
SLIDE 51

Multicast 51

AAP

  • send ACLM to claim addresses
  • object to claims and announce own via AIU
  • MAAS can preallocate addresses (ACLM) or “Adress Intent to Use” (AITU), with

reclaiming by others via ACLM

  • report periodically on address space use

September 20, 2001

slide-52
SLIDE 52

Multicast 52

MASC

  • top of hiearchy: inter-domain
  • BGP model: TCP peering relationships
  • also allows customer-provider relationships
  • send time-limited claim for range, wait a few days and then use
  • send “prefix managed” to children

September 20, 2001