SCAMP : SCAlable Multicast Protocol for Communication in Large Groups


This report proposes a new scalable multicast protocol for communication within a large group over Internet. There are many existing and new applications in which the groups involved are very large. An example would be Video-conferencing amongst a large and widely distributed group. The members of such a group can be spread across the entire Internet with no uniformity in their pattern of distribution. In such applications, there will be one or at most a few sources and a large set of destinations or receivers. The existing multicast routing protocols do not support such large groups efficiently. They incur large overheads and do not scale well. In this thesis we have proposed a new protocol that incurs less cost and is scalable.

Most protocols either build source-based trees or shared trees. Source-based trees are those which have their root located at the node sending the data. In shared trees, a few routers in the Internet act as root of the trees and distribute the data among the members of the group which are on this tree. Source-based trees have the advantage that they support high data rate. The advantage of using shared trees is that they can support sparse groups. We have proposed a protocol which combines the advantages of both type of trees. Our protocol builds both the source-based and shared trees. The shared tree is used for signalling purposes. It conveys the information regarding the source to all the members of the group. Only one signalling tree is used for all the existing multicast groups. A multicast group is referred to a set of nodes across the Internet which can be identified by a common identifier. The source-based trees on the other hand are used for data delivery. The building of the source-based tree is initiated by the receivers. It exists as long as the members belonging to that group are present. This helps in reducing the route entries maintained by routers (referred as router state). Further it reduces the concentration of traffic in different parts of the network, the processing cost of the routers and also utilizes the bandwidth efficiently, by distributing the load.

[attachment=15188]
Introduction
There are three modes of packet delivery. They are unicast, broadcast and multicast. Unicast refers to the delivery of packets from a source to one destination. Broadcast refers to the delivery of packets to all the nodes within a particular network. There is one more possibility which lies between these two cases. This is the delivery of packets from a source to a group of receivers. This is called multicasting [17]. Figure 1.1 shows the three modes of packet delivery. In this figure host HI unicasts a packet towards host H4. The packet travels the unicast path H1-R1-R2-R3-R4-R9-R11-H4. Further, host HI broadcasts the packet to all the hosts present in network 1. For multicasting it has to deliver the packet to hosts H2 and H3. The part of the tree traversed by the multicast packet is covered by routers Rl, R2, R3, R4, R5, R6, R7, R8.
With the advent of multicasting many applications have emerged in the Internet. These include Videoconferencing, Distributed Interactive Simulation (DIS) [32, 36] etc. Multicast in these applica¬tions can be either point-to-multipoint or multipoint-to-multipoint. This characterization of the appli¬cations is based on the fact they distribute data among multiple network hosts. A point-to-multipoint or one-to-many multicasting is the flow of data from a single source to many receivers simultane¬ously. In multipoint-to-multipoint or many-to-many multicasting, there are multiple sources. It can be viewed as several one-to-many data flows simultaneously.
1.1 Multicasting In IP Networks
Multicasting allows transmission of an IP datagram to a set of hosts that form a multicast group, where the members of a group can be spread across distinct physical networks. It utilizes the same best-effort delivery semantics as other IP datagram delivery.
Membership to a multicast group is dynamic, meaning that it is at the discretion of a host to join or leave a group at any time. The joining of a host to a group is subject to security (if implemented). This implies that before joining a group, the host may have to authenticate itself with some server. A host can be a member of any number of groups at the same time. If a host is a member of a particular
• Host
Figure 1.1: Modes of packet deliver
multicast group, it will receive all datagrams sent to that group. But to send a datagram to any group, it is not necessary for the host to be a member of that group. The host can be configured in three ways. Firstly, it can have no support for IP multicast. Secondly, it can have support for sending of datagrams but cannot receive multicast IP datagrams. Lastly, it can have full support for IP multicasting. In this mode, it can both send and receive the multicast datagrams.
Each multicast group has a unique multicast (class D) address. The class D address is identified by the first four bits which contain 1110. The remaining 28 bits are used for the identification of multicast groups. There is no fixed format as to how these 28 bits are to be specified.
Some IP multicast addresses are permanent and assigned by the Internet Assigned Numbers Au¬thority (IAN A). It is the responsibility of IAN A to manage the multicast address space [27, 1]. Multicast addresses range from 224.0.0.0 through 239.255.255.255. The address 224.0.0.0 cannot be assigned to any group since it is reserved. These addresses are permanent in the sense that it is the address, not the membership of the group, which always exists [15]. A permanent group may have any number of members at any time, which can be even zero. Other multicast addresses are temporary in nature and the groups which they represent are called transient multicast groups. They remain in existence as long as there are members present in the group. If the membership count falls to zero, they are discarded. Apart from this, many proposals for the management of multicast addresses have
emerged [31].
There are two types of multicast groups with regard to the distribution of the group as shown in Figure 1.2. Each router keeps information about the groups on per network basis. Within a network, there can be several members belonging to the same multicast group. But as long as there is a single member belonging to any multicast group inside the network, the router maintains information about it.