Technical Documentation

PGM Architecture and PGM Routers

PGM is defined in RFC 3208 and forms a reliable transport layer for multicast applications. Almost any multicast application can use PGM. Applications most suitable for PGM include stock market ticker update information, news reports, weather warnings, and other information that must reach multiple listeners in its entirety and in a timely fashion.

The basic PGM architecture consists of a multicast content source, one or more receivers, and zero or more routers between source and receivers. All end devices must be PGM-enabled, although there can be non-PGM routers between source and receiver. If all routers are non-PGM routers, then no routers are capable of the PGM router assistance function, and all PGM functions take place directly between source and receiver.

PGM sources send sequenced content in sessions to receivers, using multicast protocols. Other, non-PGM protocols allow receivers to learn about a particular source, its sessions, and its location. PGM receivers listen to multicast original data (ODATA), detect missing content through the sequence numbers, and send unicast negative acknowledgments (NAKs) back to the source. NAKs are answered by multicast NAK confirmations (NCFs), which suppress any NAKs from receivers on the same subnet that have not yet sent a NAK upstream. The source sends multicast repair data (RDATA) to receivers containing the missing content. PGM routers assist in this process by making sure that the negative acknowledgments follow the same path as the outbound content upstream to the source, and by suppressing duplicate negative acknowledgments and repair information.

PGM sources must maintain a sliding window of retransmittable information. There is no concept of group membership in PGM, so receivers never need to communicate with the source unless they request repair data with a negative acknowledgment. However, this means that the PGM source determines the window size for each receiver, in contrast to almost all other protocols, and requires a certain processing power in each receiver. The absence of positive receiver-to-source acknowledgments also means that PGM scales well and cuts down on control message traffic that can easily overwhelm a multicast network.

PGM receivers can start receiving a PGM session from a PGM source at any time and request any missing previous information that the receiving application needs. If the session is long enough or the transmit window small enough so that the source does not maintain a long session history, the receiver cannot get all required information.

This section describes in more detail the behavior of the three PGM elements in a multicast network:

Related Topics


Published: 2010-07-19

Help
|
My Account
|
Log Out