Related work
There are many approaches of end-to-end protocols which provide a reliable multicast transport service. Much of the existing work on this area concerns the selection of who is initiating the reliability process. There are two basic approaches to this decision: sender-initiated reliability and receiver-initiated reliability.
A sender-initiated reliability approach charges the sender with the burden of loss detection. Every receiver send positive acknowledgments (ACKs) when receiving data packets. If some ACKs are not received after a timeout interval, the sender retransmits the original data until the ACK is received (e. g., the reliable multicast transport service introduced by SCE). When the number of the members of a group increases, this approach has several limitations. First, senders must store and process a large amount of state information related to each receiver. And second, the large number of ACKs sent by receivers to each sender generates a set of simultaneous ACK-implosions, resulting bottlenecks at senders.
A receiver-initiated reliability approach charges the receiver with the burden of loss detection. Receivers send negative acknowledgments (NACKs) when detecting a lost packet (e.g., SRM, LBRM). Although this approach reduces bottleneck at senders, it presents other major drawbacks. A NACK-implosion is still possible since several receivers may lose the same packet and send redundant NACKs requesting this packet. As senders do not receive positive confirmation of reception of data from receivers, senders should indefinitely cache data for delayed NACKs. Besides, the end-to-end latency may be arbitrarily large depending on the detection process in each receiver.
Several reliable multicast protocols use a combination of these two approaches to improve performance and minimize the number of control messages exchanged between senders and receivers (e.g., RMP).
|
|
Author: Eva M. Castro eva@gsyc.escet.urjc.es |