Contents l Group Communication l Repetition: Unicast routing (brief) - PDF document

Contents l Group Communication l Repetition: Unicast routing (brief) Multicast l Multicast routing G Concepts Datakommunikation, G Intra domain ( and inter domain) påbyggnad VT-01 l (Transport protocol) Jerry Eriksson Group Communication Types of Communication l Types of Communication l Unicast (1:1) - not suitable for group communication. n*(n-1) relationships G Unicast, multicast, concast, multi peer, other l Multicast vs unicast l Multicast (1:n) l Scalability l Concast (m:1) l Applications of group communication l Multipeer (m:n) G ordering difficult G special aspects, support Unicast Multicast Sender Sender Problem Efficient one to many data Sending same data to many distribution receivers via unicast is inefficient R R Bandwidth, delay 1

Multicast vs. Unicast Scalability l Resource waste l Group size l Reliability l Not suitable for large groups G correct sequences l Time-delayed l Group awareness l Multicast is a considerable improvement l Group topology G heterogeneity Characteristics of Groups Flow control l Openness - open/closed groups l Positive acknowledgements (ACK), eg TCP l Dynamics - #members varies G Know vs Unknown group members l Negative acknowledgements (NAK) l Lifetime l Reaction of sender l Security l Transmit buffer l Awareness - who knows what? l Window-based flow control l Heterogeneity l Rate-based flow control IGMP Message types l Internet Group management protocol l Membership query: general (router send) l A host informs its attached router to join a G Query multicast groups joined by attached hosts l Membership query: specific (router send) specific multicast group G Query a specific multicast group joined by attached hosts G Operates locally l Membership report (Host send) l No network-layer multicast group G wants to join or is joined to a given multicast group membership protocol exists that operates l Leave group (Host send) among all the Internet hosts in a group G optional! (Soft state) 2

IGMP Example (2) IGMP Example (1) Membership Report Leave Group 1 3 3 1 3 Network 1 Network 2 Router Network 1 Network 2 Router 2 4 2 4 l Host 3 joins conference G Sends IGMP Membership Report message l Host 1 begins sending packets l Router begins forwarding packets onto Network 2 G No IGMP messages sent l Host 3 leaves conference G Packets remain on Network 1 G Sends IGMP Leave Group message l Router periodically sends IGMP Membership Query G Only sent if it was the last host to send an IGMP Membership Report message IGMPv3 Source Filtering Source Specific Filtering: IGMPv3 Sender 2 l Adds Source Filtering to group selection G Receive packets only from specific source R2 R1 R3 addresses Sender 1 Sender 3 G Receive packets from all but specific source addresses Senders 1, 2, and 3 are sending to the R4 If using an IGMPv2 join, router R1 same multicast group. l Benefits would forward traffic from all senders to router R4. However, in this The receiver sent an IGMPv3 Group- case with IGMPv3, no traffic from G Helps prevent denial of service attacks and-Source-Specific message to join Sender 1 is forwarded to router R4. the multicast group but to exclude all G Better use of bandwidth traffic from Sender 1. Receiver l Status: Currently an Internet Draft Unicast Routing Distance vector routing l Intra-domain routing l Distributed, iterative, asynchronous G Distance vector algorithms G Each node receives information from its directly connected nodes G Link-state algorithms G Continues until no more information is exchanged l Inter-domain routing G Nodes operates not in lock step G BGP 3

Distance vector routing (Cont) Link-state algorithms l Metric: Hops between nodes: l Global state information l Bellman-Ford algorithm l Need to know the cost of each link in the network l No complete path known l All nodes compute the routes: Dijkstra’s l Minimum costs computed and distributed algorithm l Example: RIP l Example: OSPF IP Multicast Introduction Inter-domain routing l Efficient one to many data distribution G Tree style data distribution l Routing between AS G Packets traverse network links only once l Location independent addressing l BGP: Border Gateway protocol G IP address per multicast group l Propagates path information l Receiver oriented service model G not cost information l Policy decision G Applications can join and leave multicast groups G Senders do not know who is listening • Similar to television model • Contrasts with telephone network, ATM TTL Scoping Example Scoping Multicast Traffic Receiver 1 Receiver 2 l TTL based G Based on Time to Live (TTL) field in IP header Network 2 Network 1 TTL=2 TTL=33 TTL=3 TTL=4 G Only packets with a TTL > threshold cross R1 R2 R3 R4 boundary TTL=4 Network 4 R4 blocks traffic TTL=1 Network 3 with TTL < 32 l Administrative scoping G Set of addresses is not forwarded past domain Receiver 3 Sender G More flexible than TTL based 4

Multicast Routing Multicast Routing Discussion l What is the problem? R G Need to find all receivers in a multicast group G Need to create spanning tree of receivers l Design goals Three major differences with unicast routing G Minimize unwanted traffic G Minimize router state G Scalability G Reliability Multicast vs Unicast Routing Multicast vs Unicast Routing Differences (1) Differences (2) Unlike unicast routing in which routes Multicast forwarding requires a router to changes only when the topology changes or examine more than the destination address equipment fails, multicast routes can changes simply because an application programs joins or leaves a multicast group Multicast vs Unicast Routing Differences (3) Concepts for Multicast routing l Background A multicast datagram may originate on a l Source-based routing computer that is not part of the multicast l Steiner trees group, and may be routed across networks l Trees with Rendezvous points that do not have any group members attached. l Comparison of basic techniques 5

Background Background (cont) l Group of receivers l Flooding - just broadcast l Membership of the group can change l Improved flooding - check copies frequently l Spanning trees - no loops l Network load minimized l High traffic at the root l Incremental algorithms l Multicast trees l Distributed tree is needed Data Flooding Source-based routing l Send data to all nodes in network l Receiver initiates the calculation of routing l Problem information, therefore G Need to prevent cycles G a spanning tree is created for each source G Need to send only once to all nodes in network G Could keep track of every packet and check if it had G (source, group pair) previously visited node, but means too much state l Existing unicast routing algorithms ensure efficiency G special multicast tables not necessary R2 G loops cannot occur G data can end up being sent to individual group R1 R3 members multiple times Sender Reverse Path Forwarding Example Reverse Path Forwarding (RPF) 1. Router R1 checks: Did the data Sender packet arrive on the interface with the shortest path to the l Simple technique for building trees Sender? Yes, so it accepts the 2. Router R2 accepts packets packet, duplicates it, and sent from Router R1 because forwards the packet out all other that is the shortest path to the l Send out all interfaces except the one with interfaces except the interface Sender. The packet gets sent out all interfaces. that is the shortest path to the R1 the shortest path to the sender sender (i.e the interface the packet arrived on). l In unicast routing, routers send to the Drop destination via the shortest path 3. Router R2 drops R2 R3 Drop packets that arrive from Router R3 because that l In multicast routing, routers send away from is not the shortest path to the sender. Avoids cycles. the shortest path to the sender R4 R5 R6 R7 6

Truncated Reverse Path Reverse path broadcasting (RPB) Broadcasting (TRPB) l Group membership l Small improvement to RPF G Avoids paths that do not lead to any group l Consider neighbors shortest-path to members destination l Needs information G A conventional routing table G list a multicast group reachable through each network interface l Drawback: Doesn’t consider group memberships when building the distribution tree Reverse path multicast (RPM) Summary: Source-based routing l Creates a delivery tree that spans only Technique Broad- Routing Data units Network /Multicast domain G Sub-networks with group members Flooding Broadcast All interfaces All Entire network Broadcast All interfaces from shortest path Entire network RPF G Routers and sub-networks along the shortest RPB Broadcast Interfaces on from shortest path Entire network shortest path path to sub-networks with group members TRPB Broadcast Interfaces on from shortest path Not in sub- shortest path networks without l Reflects changes in group membership group members Multicast Interfaces on from shortest path Not in sub- RPM shortest path networks without G Pruning group members Steiner trees (also for integrated circuits) Steiner trees (cont) l Establish a spanning tree that has the l NP-complete minimum overall cost. l Minimum cost is O(n log n) G Global optimization G n is the number of nodes G Cost always the same or lower than routing l Heuristics based on the shortest path G distributed solution G Cost for certain pair of nodes might be higher l Not practical l Recalculation is required whenever topology changes G Thus, a monolithic algorithm 7