One standard way of ensuring reliable delivery is introducing acknowledgments. Note how the loss of a message and the loss of an acknowledgment cannot be distinguished by the sender. That is why the messages must possess identities, and why amnesia failures, where the record of received message identities is lost, can cause repeated message delivery.

Message Acknowledgment

Another way of increasing delivery reliability is forward error correction, where redundant information is transmitted in order to increase the probability of receiving enough information to reconstruct the original message.

Two phase commit protocol ensures atomic completion of transactions. If any transaction participant fails during transaction, a rollback command is issued to all participants.

Transactional Messaging Rollback

If any transaction participant fails after commit has been decided, it must recover and proceed as directed by the coordinator.

Transactional Messaging Commit

In the multicast listener discovery protocol, the router responsible for a link periodically multicasts General Multicast Listener Queries, inviting nodes to report what multicast addresses they listen to. The nodes respond with Multicast Listener Reports, which are multicast after random delay. If a node observes a report with the same multicast address within the random delay, it drops its own report.

Group Membership Query

Protocol. 

QUERY

General Multicast Listener Query periodically multicast by router

REPORT

Multicast Listener Report multicast after random delay

A node that stops listening to a multicast address sends a Multicast Listener Drop message to the router. In response, the router multicasts Specific Multicast Listener Query to inquire about remaining listeners.

Group Membership Done

Protocol. 

DROP

Multicast Listener Drop sent from host to routers

QUERY

Specific Multicast Listener Query multicast by router

REPORT

Multicast Listener Report multicast after random delay

In sender initiated error recovery, tracking of lost messages is the responsibility of the sender. To do that, the sender requires a positive acknowledgment from each receiver on message delivery.

Multicast Sender Initiated Error Recovery

Features. 

• Can suffer from ACK implosion

• Sender must know all receivers

• Sender knows when data can be dropped

In receiver initiated error recovery, tracking of lost messages is the responsibility of the receiver. To do that, the receiver transmits a negative acknowledgment on detected message loss.

Multicast Receiver Initiated Error Recovery

Features. 

• Can suffer from NAK implosion

• Sender must transmit keepalive messages.

• Sender does not know when data can be dropped

Acknowledgment messages can be aggregated alongside the network topology.

Aggregated Multicast Error Recovery

Features. 

• Multiple variants with different acknowledgments possible

• Requires cooperation from network elements

• Can be substituted with overlay network

The Pragmatic General Multicast Protocol relies on receiver initiated error recovery with acknowledgment aggregation. Negative acknowledgments are sent to the nearest parent network element along the path, which multicasts confirmation to children along the path, possibly discards redundant negative acknowledgments and forwards to the nearest parent network element along the path again. Repair data packets are routed along reverse path created by negative acknowledgments. A node can also announce itself as a designated local repairer to a local network element. Such network element may forward the negative acknowledgments to the repairer rather than the source.

Pragmatic General Multicast

Protocol. 

SPM

Source Path Messages establish path information and perform keepalive function

ODATA

Original data packets multicast to all receivers

NAK

Negative Acknowledgment unicast to nearest parent along path

NCF

Negative Acknowledgment Confirmation multicast to children along path

RDATA

Repair data packets multicast to selected receivers

Lamport clock is a type of logical clock that reflects causality in timestamp order.

Lamport Clock

Vector clock is a type of logical clock that captures causality in timestamp values. When message transmissions are counted as significant events, vector clock can be used to provide causal ordering.

Vector Clock

The Token Ring Based Multicast Protocol provides ordering and resiliency guarantees. The current token holder is responsible for totally ordering all messages. The token rotation rules ensure resiliency.

Token Ring Based Multicast