On Thu, 14 Jun 2001, David Schwartz wrote:
Let's take an example. I'm a small provider. I have an OC3 to FooNet and an OC3 to BarNet and I get full transit on both. I sell a DS3 to someone.
Using unicast, they can consume at most 45Mbps out of my total of 2 OC3s of outbound bandwidth (which I pay for). They can consume at most 45Mbps out of my total of 2 OC3s of inbound bandwidth (which I pay for).
Well, I initially found this argument persuasive. So I did a little modeling. (Just back-of-the-envelope work.) I am now convinced that multicast is a good deal for almost everyone. 1) National Backbone - Unless the multicast group is trivially small, the revenue for the national backbone provider will be at the receiving end. The revenue is essentially unchanged. The costs are at the receiver's access link (unchanged) and in the backbone transport (significantly reduced). Good deal. 2) Small ISP's - These organizations are unlikely to host significant outbound multicast traffic. If they do have an instance of this, it is only rarely going to matter, since small ISP's are typically paying for equal capacity of inbound and outbound transit, and have outbound transit to spare. Anytime a small ISP has two multicast receivers in the network, they get to use their (expensive) inbound transit once, and bill for access twice. Good deal. 3) Regional ISP's - Well, for inbound multicast, see above. For outbound multicast a Regional ISP will have reduced backbone costs for distribution, but they will also have increased transit costs, since they almost certainly are paying for multiple transit paths. This case looks the most like the case above. If this is an ISP with a balanced access and hosting business, then hosting the multicast service will almost certainly result in some additional access revenue. Also, the regional ISP is probably limited by inbound capacity (but not always). Probably a good deal. 4) Unicast hosting provider - Now for these guys it's different. Backbone costs are minimal on outbound traffic because it gets dumped off at the nearest interconnection point. They are not likely to pick up access circuit revenue as a result of hosting the multicast, since they might not even have access circuits in their product line. They probably pay for some transit, and the capacity is limited by the outbound load. So originating multicast traffic means that the hosting company bills once and pays out every transit. Bad deal. In the long term, I'm not so sure it's even a bad deal for the hosting providers. Looking at the revenue streams in a multicast environment, the money is at the receiving end. That means that content actually is king, and these guys might get much better treatment in peering negotiations as multicast takes off.
Now I set them up with a multicast feed. They can now use up to 90Mbps of my outbound bandwidth (which I pay for). How can I charge them the same amount when it can cost me up to twice as much?
Hmm. I think the point is shallow. For example, should your transit providers charge double, too? After all, you could have to traverse their backbone east-to-west and north-to-south, but you only paid once! There might be a slight shift in some business models, since the revenue moves more to the receiving end. If you're a small ISP and you are actually limited by outbound transit, then either turn the business away or invest in the future, believing that content on your network is important and you'll pick up more access revenue. If you turn it down, someone else will want the business.