Network end users to pull down 2 gigabytes a day, continuously?
Sean Donelan
sean at donelan.com
Wed Jan 10 05:08:40 UTC 2007
On Tue, 9 Jan 2007, Valdis.Kletnieks at vt.edu wrote:
> Multicast streaming may be a big win when you're only streaming the top
> 5 or 10 networks (for some value of 5 or 10). What's the performance
> characteristics if you have 300K customers, and at any given time, 10%
> are watching something from the "long tail" - what's the difference between
> handling 30K unicast streams, and 30K multicast streams that each have only
> one or at most 2-3 viewers?
1/2, 1/3, etc the bandwidth for each additional viewer of the same stream?
The worst case for a multicast stream is the same as the unicast stream,
but the unicast stream is always the worst case.
Multicast doesn't have to be real-time. If you collect interested
subscribers over a longer time period, e.g. scheduled downloads over the
next hour, day, week, month, you can aggregate more multicast receivers
through the same stream. TiVo collects its content using a broadcast
schedule.
A "long tail" distribution includes not only the tail, but also the
head. 30K unicast streams may be the same as 30K multicast streams, but
30K multicast streams is a lot better than 300,000 unicast streams.
Although the long tail steams may have 1, 2, 3 receivers of a stream, the
Parato curve also has 1, 2, 3 streams with 50K, 25K, 12K receivers.
With Source-Specific Multicast addressing there isn't a shortage of
multicast addresses for the typical broadcast usage. At least not until
we also run out of IPv4 unicast addresses.
There is rarely only one way to solve a problem. There will be multiple
ways to distribute data, video, voice, etc.
More information about the NANOG
mailing list