The arithmetic that forces the conversation
Town hall stream ~1.5-2.5 Mbps per viewer. 3,000 on-network viewers = 4.5-7.5 Gbps of IDENTICAL content through the WAN edge. No procurement fixes that by Thursday; multiplication inside the network does:
- Peer-assisted eCDN (Microsoft's built-in eCDN, WebRTC-based): viewers fetch chunks from nearby viewers; the edge serves a fraction of the load. Zero infrastructure; needs the browser/client WebRTC path unblocked between peers and works best with dense same-site audiences.
- Caching eCDN (partner appliances/software): explicit caches serve each site; deterministic, network-team-friendly, more to deploy.
Both integrate with town halls/live events via tenant configuration (Microsoft eCDN is a toggle + licence question; partners via their connector). Streams are delivered over HTTPS — eCDN multiplies delivery, it does not change the protocol story.
The pre-event proof (non-negotiable)
- Pilot event to a representative site with eCDN active.
- Measure: edge egress for the event window vs viewer count — the offload ratio (eCDN analytics report it; the WAN graph confirms it).
- Chaos-check: one viewer VLAN where peer traffic is blocked — confirm fallback to direct streaming works (eCDN failures must degrade, not black- screen).
An unproven eCDN before a CEO event is a rehearsal you skipped.
What to watch (proofs)
- Offload ratio during events: eCDN analytics (Microsoft eCDN dashboard / partner console) — % served peer-to-peer or from cache; the number the WAN team frames.
- The edge graph: WAN egress during the event vs viewer count — offload realised (flat edge, thousands of viewers) or not (linear climb = eCDN silently off).
- Client-side verification: Microsoft eCDN offers a per-client debug/analysis view — sample viewers at each big site during the pilot.
- Config drift: the events policies + eCDN settings in your nightly scan — the toggle reverting after a licence lapse is a drift alert you want BEFORE the all-hands.