The authors examine the performance implications of providingreliability in conjunction with multicast transport over a high speedwide area network. They use a block based acknowledgement and selectiveretransmission protocol to evaluate the impact of the loss rate and themulticast tree topology on the achievable throughput. Their results showthat even when the buffer overflow probability at switches and receiversis low, the cumulative loss probability seen by a source may be quitehigh. They also demonstrate that the average throughput increasessignificantly if the transport protocol delivers packets to theapplication layer out-of-sequence. They investigate the scalingproperties of the error control mechanism and show that the multicasttree topology that results in minimum transfer time is not necessarilythe same as the one constructed using minimal bandwidth or shortest pathalgorithms
展开▼
