There has been abundant debate lately about Comcast's impending implementation of a 250GB/Month bandwidth cap on it's users in the United States. Time Warner had started capping in June of 2008 by limiting subscribers to 40GB/Month, and many international ISPs regularly implement bandwidth limitations for their users.
The reason behind this is simple, the majority of providers of Internet access come from a metered-service background, be they telephone calls or cable channels. Now they want to apply the same model to Internet Access. This is going to work out badly for many emerging Internet application classes.
During the go-go days of the dot-com boom, I was building remote access servers that allowed Internet service providers to provide modem and ISDN dial-up access to the Internet. They would take twenty or so of these boxes, each of which provided between 48 and 60 56k modem connections or 64k ISDN connections, and aggregate them together with a backhaul router as a 'point-of-presence'. What no one told you was that they would take all of these access servers, each capable of driving 2 T1/E1 of traffic (3-4Mbps), and route them through a fraction of the backhaul bandwidth. A common metric that we would see was 20:1, although there was one Asian ISP that regularly provisioned 100:1. That means for every 20 ports, there was one port worth of Internet backhaul bandwidth. If two people got on, then divide that bandwidth by two, and so on.
This practice of oversubscription has a long history in the telephony business by the use of probability distributions in provisioning, commonly referred to as 'Erlang tables' (in reference to the Danish mathematician A.K. Erlang). This is the valid assumption that of, say, 100 phones in your business, only a certain percentage of those will be off-hook at a particular moment and a subset of those will be placing calls to the PSTN, which in turn require an outbound DS0.
When you have explicit 1:1 relationships between telephone calls and DS0s, these Erlang calculations make perfect sense to ensure that you do not over-provision your bandwidth and experience the dreaded 'fast-busy-signal'. When you have constantly updating Internet devices like the Chumby, Ambient Orbs, LG BluRay/NetFlix players, and computer-based RSS-readers, it loses it's explicit 1:1 relationship and becomes more of a free-for-all.
The ISPs want to ensure that their time-tested oversubscribing practices are allowed to continue unhampered, with their financial benefits displayed proudly in the Operational Expenditures column. They are regressing to the good old days of metered service, but for broadband connections.
This will have a chilling effect on streaming media sites like Hulu and NetFlix's on-demand service, who depend on the traditional unlimited bandwidth model of Internet access. It will also have broad architectural ramifications for social networking sites, games, and serious virtual worlds that encourage user-created-content and rich streaming. If I ship you a DVD of video/audio assets, like Little Big Planet or Warcraft, then the network updates are limited and you are drawing assets from your local drives. If it's predominantly user created content and streaming, then I have to ship you said content over the network along with updates on other avatar locations and voice/chat/video. This consumes bandwidth.
Just using Second Life as an example, if you consider a user that isn't just standing with two or three other people chatting but actually decides to leverage the multi-modality of the medium with audio chat, video streaming, perhaps some collaboration, while also moving around more than once in blue moon, you end up with between 200-300kbps of traffic. If you are using SL as your primary collaboration tool for the work-day, that adds up to around 25GB of traffic a month, before you add in your email and other network services. Obviously, adding bandwidth caps to consumers will favor the non-user-created-content model of virtual world architectures, else a ton of caching from an expanded palette of content primitives.
Lets hope some cooler minds prevail at the ISPs, or the customers are vocal enough in their distaste for bandwidth caps. I can empathize with the ISPs who see that 80% of their bandwidth is consumed by P2P traffic coming from a small percentage of their users, but we all know that individual (non-bit-torrenting) network traffic is on a not-so-subtle upward trend year over year, so any arbitrary bandwith limit is bound to be a bucket of cold water on application innovation on the Internet.
Comments