only to B and let B forward it to A (Fig. 1c), halving its bandwidth usage. B now has to additionally send C’s updates to A, but B’s overhead in this situation is usually lower, because it can piggyback C’s messages to its own, assuming there is some timing synchronization of B’s and C’s updates. Especially when the message payload is small, saving packets, and thus packet headers, reduces the total traffic. Compression and application-specific message aggregation techniques can bring further savings. But even if the total traffic reduction is marginal, this option allows shifting traffic between nodes. If B has more spare bandwidth than C, this may still be a good choice. In any case, the rerouting has to be paid with an increased latency on the path from C to A.1 The decision whether to use the indirection depends on the available bandwidths and latencies between the nodes and finally on the application requirements.