Broadcast television in an IP world

William Herrin bill at
Mon Nov 20 19:22:19 UTC 2017

On Fri, Nov 17, 2017 at 7:48 PM, Baldur Norddahl <baldur.norddahl at>

> Does multicast have any future?

Multicast is a fine replacement for local-lan (i.e. direct connected
interface) broadcast.

For video distribution, multilevel caching simply works better. It's no
deep mystery why.

Wide scale multicasting requires ISPs to allow the most critical inline
resource (the routers) to accept fine-tuned instructions from any third
party that wants to be a content head-end. For ordinary routing we don't
allow anything more fine tuned that a /24 and we've been reluctant to allow
that. We're somehow going to beef up the routers to allow non-paying third
parties to fine-tune down to the video stream? Not happening.

Multilevel caching consumes upwards of an order of magnitude more data
transfer than an optimal multicast system, but it does it to the side, out
of the critical path. If a node breaks, it doesn't take down the network
with it. And you can slap a cheap server in place

And of course caching supports time-shifting which multicast does not.
Given how people consume video today, that's an important distinction.

Peering in to my crystal ball, I see an abstracted caching system which
isn't tied to any particular vendor. Fetch decryption keys directly from
the video vendor, then find the nearest cache via a solicitation to a
protocol-standard anycast IP address. The cache fetches from the next cache
up. ISP deploys as many caches as it finds convenient and cost effective.
Paid megacache at the top of the hierarchy located at the carrier neutral
data center that's cheaper than transit for both the eyeball networks and
the content providers. Data cached in chunks of arbitrary data that only
mean anything to the producer and consumer but are identified in such a way
that multiple consumers for the same data request the same chunk ID. And
small enough chunks that real-time feeds are delayed by few enough seconds
to make them practical.

Unicast with a little bit of anycast. No multicast on that road map.

Bill Herrin

William Herrin ................ herrin at  bill at
Dirtside Systems ......... Web: <>

More information about the NANOG mailing list