On Jan 20, 2006, at 9:29 AM, Peter Corlett wrote:
<Michael.Dillon@btradianz.com> wrote: [...]
But it's no magic bullet. Streaming live media also requires low jitter, especially if you are selling it as TV because viewers will join and leave channels often, as they change channels on their remote controls. This means you can't have big local buffers to hide jitter, therefore you have to build a network with enough capacity so that packets are all cut-through switched.
I observe about 3-4 seconds of latency on the UK DVB-T and DAB broadcasts anyway compared to analogue. Cost-cutting on CPU grunt in decoder boxes can mean it takes up to ten seconds to change channel.
<AOL> Here in the US, Comcast's "digital cable" service takes seconds to show a picture after you change channels. I don't know if that's buffering or CPU or what, but consumers are clearly OK with it. </AOL> So you _can_ have a large client-side buffer and ignore jitter. That means packet loss is important, not jitter. (A 2 second buffer would be orders of magnitude more than your typical jitter.) Which means queue size is only relevant if you drop things off the back end of the queue. Which means you can build an intentionally congested network and "sell" the front-end of the queue to services which will pay you more. The rest will just risk being dropped off the end of the queue. Will consumers care? Hell, they're already used to the Internet not really working right, rebooting their computers every day, and sites being taken down 'cause the next box over is infected and DDoS'ing someone (or their domain has been removed for spamming :). In fact, most consumers probably can't use the speed they have since their computer is using all the available bandwidth & CPU spewing crap onto the 'Net from the 1389 viruses installed. So, yeah, I think the end user will put up with the fact some sites are slower on their DSL line and not look to change providers. And they will slowly migrate to the faster sites - i.e. the ones who pay for the front of the queue. Also, no one has talked about the ideas proposed in Vixie's second link: That the big content providers are willing to pay a 'little' to raise the bar of entry. A few million bux a year to each of the RBOCs in the US would be a rounding error in Google's bottom line, but it would make it nearly impossible for a 'start-up' to make it. Doesn't that scare anyone?
In contrast, streaming video and audio from iTMS starts to play a lot quicker. It sounds like the problems with jitter and latency over private IP networks is overstated if it still works fine over the Internet.
(FWIW, this is on 1Mb/s ADSL that is 170ms from www.apple.com.)
Yeah, but you don't get iTMS stuff from www.apple.com. I'm betting you are a LOT closer to iTMS. :-) -- TTFN, patrick