Reading for Lecture Nov 20

A Reliable Multicast Framework for Light-weight Sessions and Application Level Framing

In the IP multicast, data sources simply send to the group’s multicast address without any advance knowledge about the group’s membership. To receive data, receivers simply announce that they are interested. Each receiver joins and leaves the group individually, without affecting the data transmission to any other member.

The SRM maximizes information and data sharing among all the members, and strengthens the individuality of membership by making each member responsible for its own correct reception of all the data. SRM design follows the core design principles of TCP/IP. It requires only basic IP delivery model – best effort with possible duplication and reordering of packets – and build the reliability on an end-to-end basis. Furthermore, SRM algorithms dynamically adjust their control parameters based on the observed performance within a session.

TCP multicast style has a number of problems like if all receivers send ACK, the sender will be overwhelmed by this message. Also, how does a sender continuously track the active receivers and the reception rate of each. Round trip time is also very difficult to estimate due to the difference in propagation time to different receivers.

In the Wb framework, each member is identified by a globally unique identifier, the Source-ID and each page is identified by the Source-ID of the initiator and the page number locally to that initiator. The wb multicast design assumes: 1) All data has a unique name. 2) The name always refers to the same data. 3) Source-ID are persistent. 4) IP datagram delivery is available. 5) All participants join the same multicast group, no distinction between the senders and receivers.
When a receiver detects missing data, it will send a repair request at a random time.

Scalable Application Layer Multicast

This system is based on a hierarchical clustering of the application-layer multicast peers and can support a number of different data delivery trees with desirable properties. The application-layer multicast protocols can be evaluated along three dimensions: 1) Quality of data delivery path. 2) Robustness of the overlay. 3) Control overhead.

The project aims at developing an efficient, scalable, and distributed tree building protocol, which did not require any underlying topology information. This reduces the worst-case state and control overhead at any member to O(logN). An average member maintains state for a constant number of other members, and incurs constant control overhead for topology creation and maintenance.

Hosts a distributed in different layers with following properties: 1) A host belongs to only a single cluster at any layer. 2) If a host is present in some cluster in layer L_i, it must occur in one cluster in each of the layer, L_0, …, L_i-1. In fact, it is the cluster leader in each of these lower layers. 3) If a host is not present in layer L_i, it cannot be present in any layer L_j, where j > i. 4) Each cluster has its size bounded between k and 3k-1. 5) There are at most log_kN layer, and the highest layer has only a single member.