The explosive growth of one-way data flows in modern interconnects, now routinely in the \ (10^10–10^12 \) b/s (100 Gb–1 Tb) range, shows no sign of slowing. Yet, while one-way throughput scales, two-way (acknowledged) communication remains fundamentally bounded by the round-trip speed-of-light latency. This contradicts assumptions in many network architectures that model performance and congestion control only in terms of one-way delay. A crucial shift emerges when Ethernet frame lengths exceed the physical length of the underlying link. In this regime, acknowledgments of each frame occur during transmission and incur almost no penalty, enabling a reconceptualization of classical Shannon theory. By reversing the time-oriented Turing tapes at both ends of the link and comparing sent and received bits with hardware comparators on the SERDES interfaces, one-way entropy analysis can be generalized into a two-way Shannon channel. This reframing directly integrates feedback into the definition of information itself. The implications extend beyond communications engineering. At the conceptual level, this analysis resonates with well-known debates in the foundations of physics, especially the interplay of information, entropy, and time symmetry. We propose a new model of temporal structure, termed Alternating Causality, which formalizes time as a reversible bidirectional process, and information as a conserved quantity.
Munamala et al. (Wed,) studied this question.