MULTICAST ROUTING PROTOCOL COMMUNICATION NETWORKS 2. Draft Document version 0.3

Transcription

1 MULTICAST ROUTING PROTOCOL COMMUNICATION NETWORKS 2 Draft Document version 0.3

2 Table of Contents 1 INTRODUCTION: Overview Terms and Definitions Conventions Assumptions Layer 1 and Layer Topology Addressing Scheme Example NEIGHBOR DISCOVERY Procedure Packet Format TYPE field Node ID Example Reliability ROUTING AND FORWARDING Routing Table Construction Discovery Routing Update Packet Format Periodic Routing Updates Reliability Forwarding Data message Format Type Field Sequence Number Destination Addresses Example Algorithm for multicast routing Algorithm for message forwarding Reliability State Diagram RELIABILITY SCHEME Reliability Time-out Procedure Packet Format of ACK TYPE field Sequence Number Example APPENDIX A EXAMPLES OF MULTICAST ROUTING

3 1 INTRODUCTION: 1.1 Overview The purpose of this document is to provide a standard for the implementation of small multicast routing in a small to medium sized network. This document includes a description of the procedures, packet formats and algorithms to support the following functions and features: 1. Neighbor Discovery by all nodes upon bootstrapping and periodically thereafter. 2. Routing protocol to exchange end-node information among routers and build a routing table. 3. Algorithm for forwarding uni-cast and multi-cast packets toward the destination using the routing table. 4. Reliable Message Delivery. 1.2 Terms and Definitions Duplex: A type of connection between two nodes whereby both of them can communicate with each other at the same. Host: Machine used by the user to send/receive messages. Also referred to as end-node in the document. Multi-casting: Sending data packets to multiple destinations. Neighbor node: A node directly connected to the node under consideration. Node: Host or Router Packet: Similar to message. Packet and message can be used inter-changeably. Port: A connection identifier Reliability: Ensuring that the message sent has been received by the receiver. Router: Machine that serves to route packets between hosts. Routing Table: A table maintained to keep track of all hosts in the network. Uni-casting: Sending data packet to a single destination 3

4 1.3 Conventions In this document SHALL, SHALL NOT, MUST and MUST NOT express a mandatory requirement while MAY expresses an optional feature. 1.4 Assumptions Layer 1 and 2 1. All links in the network are unreliable with a packet loss probability of Packets may get lost but cannot get corrupted. Hence error detection is not required. 3. Packet size will always be with in the Maximum Transmission Unit (MTU) size and hence fragmentation and re-assembly of packets is not required. 4. Links failure will not occur. 5. Link delay will be 100ms. 6. All nodes have a Duplex connection between them Layer 3 1. There is no separate control port used by routers. Control signaling and data packets share the same one duplex port. 2. Multi-cast packet is sent from sender to 1-3 recipients. 3. Data packets do not need to be delivered in order Topology 1. Routers have 2-4 ports through which they connect to other routers and endnodes. 2. It is possible that network entities may join the network. But the network entities already existing in the network do not fail. 3. Each end-node has a single direct connection to the router. 4. Direct links are not present between end-nodes. 5. Topology establishment is pre-specified by designating the ports and their connection relationships. The protocol may not need to support dynamic topology setup. However, this global topology information is not known by any network entities a priori. 6. There will be at least 5 hosts and 5 routers in the network. 4

5 1.5 Addressing Scheme Routers and Hosts share the same fixed addressing scheme. 1 byte MUST be reserved for the Router/Host ID. ASCII values less than or equal to 127 SHALL be reserved for hosts whereas those greater than 127 SHALL be reserved for routers Example In the above example, hosts, denoted by squares, have IDs <=127 and routers, denoted by circles, have IDs > NEIGHBOR DISCOVERY 2.1 Procedure 1. When the node (router/host) boots up, Discovery routine is called. 2. Discovery packet, described in section 2.2, SHALL be transmitted on all outgoing ports every 400ms. 3. At the router receiving port : a. If the received packet is TYPE 1, then packet type is hello message b. Following 8 bits SHALL then be taken to be the ID of the node connected at the other end. c. ID <= 127 implies that a host is connected at the other end. In this case routing table is updated to contain that host entry, if that host is not already present in the routing table. d. If ID>127, hello message is ignored. 4. Continue Discovery phase until the node is off. 5

6 Both hosts and routers run this protocol at their end. Hosts discard hello messages received from routers. 2.2 Packet Format The general packet format is shown below: TYPE NODE ID 1 byte 1 byte TYPE field Type field SHALL be used to differentiate between different kinds of packets used in the network. ACSII value 1 MUST be used in the type field of Hello messages used for neighbor discovery Node ID Node ID MUST follow the addressing scheme mentioned in section Example 1. Hello Message from Router byte 1 byte Here the node ID is greater than 127 identifying the node as a router. 2. Hello Message from Host byte 1 byte Here the node ID is less than or equal to 127 identifying the node as an endnode/ Host. 6

7 2.3 Reliability The neighbor discovery procedure SHALL be unreliable which means that a hello message with TYPE field set to 1 does not need to be acknowledged by the receiver. 3 ROUTING AND FORWARDING Routing in the router network corresponds to selecting the ports out of which an incoming message will be forwarded for each of the destinations included in the message. If the same port is selected for more than one destination, then the message MUST be sent as a single packet for these destinations; in other words, one packet MUST be forwarded for each distinct port selected and the packet header MUST be appropriately modified to contain only those destinations which should be reached by that particular port. Port selection is based on minimizing the maximum possible total number of hops for forwarded packets on the path(s) starting on the selected port(s), where the total number of hops is calculated as the sum of single-packet transmissions over each link in the network until each packet is delivered to its destination. In order to do this, each router MUST keep a record of the shortest distance to every possible destination (end-node) out of every one of its ports. This information MAY be kept in a matrix or similar data structure, referred to as a routing table, indexed by destination and port number. 3.1 Routing Table Construction Routers SHALL populate their routing tables by listening for updates from their neighbors. These updates SHALL be of two types Discovery The first type corresponds to discovery messages received by routers from neighboring end hosts, which will contain the host name. When a discovery message from a neighboring end host is received, the update to the routing table MUST assign a distance of 0 to the host name received in the discovery message at the port through which the discovery message was received. Details of discovery including the packet format are discussed in section 2 of this document. 7

8 3.1.2 Routing Update The second type corresponds to routing table updates received from neighboring routers, which have the format shown in section When this type of message is received on a port port_in, the router MUST execute the following procedure: Algorithm UpdateRT (in: INT port_in, in: SET update) Preconditions: SET PORTS: Ids of routers ports (e.g. {1, 2, 3, 4}) MATRIX RT[n PORTS ]: Routing table that contains the distances of the shortest paths to every host out of every port (n is the number of hosts, and x denotes the number of elements in data structure x) SET update contains (host, distance) pairs received in a routing update Postconditions: RT is updated with the minimum distances to the hosts in the update, if the distances received are less than the existing ones. BEGIN for each (host, distance) update do if (RT[host, port_in]=nil OR distance < RT[host, port_in]) RT[host, port_in] = distance + 1 end if end for END 8

9 Packet Format TYPE # of hosts HOST1 DIST1 HOST2 DIST2 HOSTn DISTn Figure 1. Packet format for routing table updates. Packets contain variable number of hosts specified in the # of hosts field and distances that correspond to the distances of the shortest paths to those hosts from the updating router Type Field For Routing Update messages, Type field MUST be set to ASCII value Example In the example topology of section 1.4.1, Router 134 might send the following message to router 133. (This would be the case when router 134 has received information about Hosts 2, 3 and 4 but not about host 1) Periodic Routing Updates After a router has received its first update (of either type), it MUST send periodical routing table updates out of all ports connected to other routers. Updates MUST be sent at an interval of 2 seconds using the packet format shown in section A router 9

10 MUST NOT send updates if it has no information in its routing table. A host receiving a routing update will discard it. The content of a particular update to be sent from a particular port update_port is given by the following algorithm: Algorithm CreateUpdate (in: INT update_port, out: SET min_distances) Preconditions: SET PORTS: Ids of routers ports (e.g. {1, 2, 3, 4}) MATRIX RT[n PORTS ]: Routing table that contains the distances of the shortest paths to every host out of every port (n is the number of hosts, and x denotes the number of elements in data structure x) Postconditions: min_distances contains the minimum distances in RT to each host over all ports, excluding the port on which the update will be sent. BEGIN min_distances = for each host RT INT distance = min port PORTS {update_port} (RT[host, port]) min_distances = min_distances {(host, distance)} end for END Once the set of minimum distances is determined, the router constructs a message using the format shown in section , with # of hosts field set to min_distances and Host data (Host ID and Distance) set to each one of the entries in this set, and sends the message from update_port. The router MUST repeat the above procedure for each port to which a neighboring router is connected Reliability The routing updates SHALL be unreliable which means that a node receiving a packet with TYPE field set to 2 does not need to acknowledge it. 10

11 3.2 Forwarding Data message Format DATA Messages exchanged between hosts through the router network MUST follow the format shown in Figure 2. TYPE SEQ NUM # OF DEST DEST1 DEST2 DEST3 LENGTH PAYLOAD 1 byte 1 byte 1 byte 1 byte 1 byte 1 byte 2 bytes variable (max 1000 bytes) Figure 2. Packet format for end host messages. The message may include 1 to 3 destination ids (DESTi), indicated by the #DEST field. These destinations are followed by a variable length (in Bytes) payload, indicated by LENGTH. The maximum length of a packet is 1000 bytes Type Field For Data messages, Type field MUST be set to ASCII value Sequence Number The Sequence number is used in the data packet to avoid reception of duplicate packets at the receiver due to the reliability scheme implemented at the link layer. Any node receiving the data packet SHALL send back an ACK (described in section 4.4) with sequence number set to next expected sequence number at that port. Any node forwarding a data packet at a port SHALL set the sequence number to that received in last ACK at that port. If no ACK has been received at that port since boot-up, the sequence number shall be set to zero. The sequence number SHALL go from 0 to 255 before wrapping around. NOTE: Sequence numbers are specific to links. Thus a router having four active links would need to keep a track of four separate sequence numbers, one at each link Destination Addresses The DEST2 and DEST3 fields are optional depending on the number of destinations in the # OF DEST field. Hence, we would have a variable length header. 11

12 Example TYPE =3 Seq # =12 # of DEST = 2 DEST1 = 2 DEST2 = 4 LENGTH = 800 PAYLOAD 1 byte 1 byte 1 byte 1 byte 1 byte 2 bytes 800 bytes Algorithm for multicast routing With the routing information, each router SHALL implement the RouteMessage algorithm and execute it to select the port(s) upon reception of a message on an incoming port. The interpretation of the local variables used in this algorithm is as follows: P: Set of all possible combinations of i ports, where i is the number of destinations to which the message is addressed. The number of destinations supported by the algorithm MUST be at least 3. maxhops: Holds the maximum number of hops on the paths that may result from sending messages out of a particular port combination. distinct_ports: Holds the number of distinct port identifiers in a particular port combination for a number of destinations (e.g. for these port combinations (1,2,3), (1,2,1), and (3,3,3), the value of distinct_ports is 3, 2, and 1, respectively). shared: Holds the number of destinations that can share the port of some other destination in a particular combination of ports, obtained from the number of destinations and distinct ports for them. minmax: Holds the minimum of all maxhops values explored by the algorithm for the different combinations of ports up to a certain point. Upon termination, it will hold the overall minimum value. ports_shared: Holds the maximum value of shared variable for the port combinations whose maxhops value is minimum. Algorithm RouteMessage (in: LIST destinations, in: INT port_in, out: LIST ports_out) Preconditions: SET PORTS: Ids of routers ports (e.g. {1, 2, 3, 4}) MATRIX RT[n PORTS ]: Routing table that contains the distances of the shortest paths to every host out of every port (n is the number of hosts, and x denotes the number of elements in data structure x) Postconditions: ports_out contains an ordered list of port identifiers, one for each destination 12

13 BEGIN SET P = {(x 1,,x i ) i = destinations x i PORTS-{port_in}} INT minmax = INT ports_shared = 0 for each (p 1,,p i ) P do INT maxhops = RT[destinations i, x i ] SET distinct_ports = {x i i {1,, destinations }} INT shared = destinations - distinct_ports maxhops = maxhops shared if (maxhops < minmax (maxhops == minmax shared > ports_shared)) minmax = maxhops ports_shared = shared ports_out = (p 1,,p i ) end if end for END Exceptions: If the routing table contains no entry for a particular destination i, then entry i must be eliminated from destination and ports_out lists Algorithm for message forwarding The output list ports_out MUST have the same order as the list of destinations, so that the packet addressed to destinations i will be sent out of ports_out i. The following procedure is used for forwarding messages given the result of the RouteMessage algorithm: Algorithm ForwardMessage (in: LIST destinations, in: LIST ports_out, in: message) Preconditions: ports_out has been obtained using the RouteMessage algorithm for the given destination list. Postconditions: The message has been sent to the given destinations from the correct ports 13

14 BEGIN SET distinct_ports = {ports_out i i {1,, ports_out }} for each p distinct_ports do SET port_destinations = for each i {1,, ports_out } do if (ports_out i == p) port_destinations = port_destinations destination i end if end for Create packet (Figure 2) with: #DEST = port_destinations for each d port_destinations do Add destination d end for LENGTH = message Add message end Create Send packet out of port p end for END In the above procedure, distinct_ports has the same interpretation as in the RouteMessage algorithm. Set port_destinations is used to aggregate all destinations to be sent out of a particular port, and will be used to address the packet to be forwarded from that port only with those destinations. A router MUST NOT forward any packet corresponding to a message on the port at which this message was received (port_in above). Correct implementation of the routing algorithm will ensure that this does not happen Reliability Data messages SHALL be reliable which means if a receiver receives a packet of type 3, it MUST send back an acknowledgement. 3.3 State Diagram Figure 3 shows the state diagram for routers. Rectangles denote states while ovals denote procedures, some of which correspond to the algorithms described above. Bracketed transition labels indicate an input to a procedure or state; otherwise, labels indicate the actions that trigger state changes. 14

15 Startup Update Arrives Update RT Message Arrives Update Arrives [RT] Ready Clock Trigger Message Arrives Create Updates Route Message [min_distances] [ports_out] Send Updates Forward Message Figure 3. Routing State Diagram. 4 RELIABILITY SCHEME Stop-and-Wait ARQ SHALL be used for reliability. 4.1 Reliability Acknowledgement (ACK) packet will be sent by the receiver after receiving the packet. 4.2 Time-out The Stop and wait retransmission time-out is set to 250 ms. 4.3 Procedure 1. Whenever a node (Router or Host) receives a packet with TYPE field set to 3 (data packet), it MUST send back an acknowledgement (ACK) packet described in section 4.4 with TYPE set to If the sender does not get the ACK packet with-in 250 ms, sender will time-out and will retransmit the Data packet. It will keep on doing so until it gets the ACK packet. 3. If the receiver receives a duplicate packet, it will discard the packet and retransmit the ACK. 15

16 4.4 Packet Format of ACK TYPE SEQ NUM 1 Byte 1 Byte TYPE field For the acknowledgements, the TYPE field MUST be set to ASCII value Sequence Number The sequence number of the ACK SHALL be set to next expected sequence number at that port Example Following ACK would be sent in response to the data packet described in section TYPE = 4 Seq Num = 13 1 Byte 1Byte 16

17 APPENDIX A EXAMPLES OF MULTICAST ROUTING Host Port When sending a message received from 1 to ['2', '3', '4'] the steps below show how the ports or the minmax value change. Set minmax to 10 from ports 1, 1, and 1. Set minmax to 9 from ports 1, 1, and 3. minmax is 9 going out of ports 1, 1, and 3. Sending to [ 2, 3 ] on port 1 and [ 4 ] on port 3. 17

18 Example of application of the routing algorithm with a given routing table and different message destinations: Host Port 0 1 Inf Inf Inf Inf Inf 4 5 Inf Inf Inf These show the intermediate steps where minmax is set or the ports are changed and which port are selected for which messages in the end: minimax('4', '5', '6') Setting minmax to 10 from 0, 0, and 0. Setting minmax to 9 from 0, 0, and 2. minmax is 9 going out of ports 0, 0, and 2. Sending to [ 4, 5 ] on port 0 and [ 6 ] on port 2 minimax('1', '2', '3'); Setting minmax to 11 from 0, 1, and 1. Setting minmax to 9 from 0, 1, and 2. Changing ports to 0, 3, and 3. minmax is 9 going out of ports 0, 3, and 3. minimax('5', '6'); Setting minmax to 9 from 1, and 0. Setting minmax to 8 from 1, and 1. Setting minmax to 7 from 1, and 2. minmax is 7 going out of ports 1 and 2. Sending to [ 5 ] on port 1 and [ 6 ] on port 2. 18