Dear Valdis; On Jan 9, 2007, at 10:02 PM, Valdis.Kletnieks@vt.edu wrote:
On Tue, 09 Jan 2007 11:29:32 EST, Gian Constantine said:
If you considered my previous posts, you would know I agree streaming is scary on a large scale, but unicast streaming is what I reference. Multicast streaming is the real solution. Ultimately, a global multicast network is the only way to deliver these services to a large market.
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?
This is a very good point. It is very reasonable to expect viewing choices to follow a Pareto distribution (such as Zipf's law or the 80-20 rule). That plus some reasonable economic assumptions make 30K commercial channels not an unreasonable assumptions in a few years. But that also implies that it is _not_ realistic to have "30K multicast streams that each have only one or at most 2-3 viewers." You may have 30K streams, most may have only a few viewers, and still have fairly large savings. To flesh out your example, if you have 1 million viewers on your network, and if you assume 30K channels and the same Pareto distribution as web sites, - the largest channel has 1.8% of the audience - 50% of the audience is in the largest 2700 channels - the least watched channel has ~ 10 simultaneous viewers - the multicast bandwidth usage would be 3% of the unicast. These same models IMHO makes cell phone RF multicast not incredibly compelling. Because there is less feedback on a multicast RF, the power has to be greater for a multicast channel, and in that case the bandwidth (or RF power) savings are small or even negative, because you only have maybe 100 people on a cell watching. Regards Marshall