Live streaming is the workload that exposes every weakness in a delivery architecture, and it does so in public. Traffic does not ramp; it arrives. Content cannot be pre-warmed at the moment it matters, because the segment a viewer needs was created two seconds ago. And the audience judges the result within a few seconds of pressing play, with no patience for an explanation about a congested peering point in one metro.
For most of the past decade, the answer was to buy a bigger network. That answer has weakened as live audiences have grown more global and more concurrent. No single content delivery network is the fastest everywhere at once; capacity is provisioned regionally rather than in infinite supply, and the failures that ruin live events are usually partial rather than total. What ...
