A. Because it indicates a cable failure.
B. Because in synchronous parallel processing (like MPI or NCCL All-Reduce), the entire job must wait for the slowest node/packet to arrive before proceeding to the next step.
C. Because high tail latency causes the CPUs to clock down to save power.
D. Because tail latency determines the maximum storage capacity.