-
Why (Problem Statement)?!
- With large live audences and moar bitrates, are we at inflection
point?
-
Live streaming != On-Demand Streaming
- Expectations for low latency. If you hear the bar downstairs
cheering before you see the person with the ball do, um,
whatever people do with balls in sprts things..... Clearly
the minute taker doesn't sport...
-
Join rates are vastly differnt; smooth for on-demand, step
function for live
-
Multicast has been successful in some places.
- Not so wonderful on the Internets...
- "All or Nothing" -- all L3 hops need to do this.
- "Too Complex" -- perceived benefit not worth the cost.
(WK/scribe comment: And often the benefit is negative..)
- "Chicken and Egg" -- no-one does this, because no-one does this.
-
TreeDN...
- Native and overlay concepts to deliver serive to users who don't
have multicast.
-
Native On-Net SSM
-
Overlay - AMT (RFC7450)
- Dynamically built tunnels hop unicast-only bits of the
network.
- Solves "All or Nothing" and "Chicken & Egg"
-
Incremental deployment
-
Diagrams (with and without Multicast and TreeDN)
- If deployed on existing infrastructure, it is basically
CDN-on-a-Chip - $0 capex, ~$0 opex
-
Benefits
- More efficent network
- Allows SP to offer Replication As A Service
- Democratizes and decentralizes content sourcing
-
Applicability
- Any multi-destination content
- Live treaming is the sexy example
- But large files (e.g sw updates) also work well...
-
Next Steps / Action Items:
- Q (Kyle): Read draft, interesting... Still had issues, like
democratizing content - this is just for the fanout though, not
the rights issues, etc. Creating novel attack surface with
replication as a service. This hand-waves away things like key
mgmt - they are still there, just not in the replication.
Exposes what content is being consumed. Willing to help, but
still unresolved (non-technical) issues.
- A: 1: Yes, content rights are an issues, but thats orthogonal -
e.g nature cams, drones, etc. 2: Attacks: Multicast is a
differnt attack surface. Not more or less, just differnt. 3: Key
Mgmt - What was meant was the key mgmt is that it goes on
between user and provider; intervening providers don't have to
participate. Not unique to mulicast...
- Q (Eric Vynke): Looks like this is best current practices for
using MBONE best current practices "just" stitching existing
bits together, not new work. Seems like it belongs in MOPS. Is
there a deployed PoC?
- A: Yes. Come to MBONED...
- Q (Hang Shi): What transport does this use? Can be used with
TCP?
- A: Nope. Multicast doesn't really do TCP, works with UDP - can
carry any multicast compatiable transport.
- Q - Chairs: WG Adoption? If so, MOPS or MBONED? Show of hands...
- Chairs: Looks like there is good support for adopting
(approx 23 hands raised, 3 not raised in Meetecho queue)