On Tuesday, December 21, 2010 11:26:48 am Rettke, Brian wrote: > The problem is probably not the connection speed, but congestion on the CMTS. > If the downstream is saturated (too many people watching Netflix on a node) > the available shared bandwidth may not be enough to support your real-time > traffic. Which is a pretty good archetype for the discussion anyhow.
Well, at the time we did this test NetFlix was still just a DVD by mail outfit; this has been a couple or three years ago. Congestion == oversubscribed. I would love to see a public posting or notice or something on my ISP's website showing current flows and congestion (the Cacti driven Network Weathermap is one such tool I've seen networks use; one of my providers used to have one publicly available, and it was very useful). Would make it much easier to make informed decisions on my part. But this CMTS subscriber wanting to do medium-low bandwidth H.323 never had trouble seeing our stream to him; that was the funny thing. It was always the return stream from him to us that broke up. And it didn't act like congestion; it acted like some sort of filter in place that would only allow the full upstream briefly, and then would die for some period of time, and then would allow another burst of traffic. (I've received one private reply mentioning a possible technology to do this....) Many if not virtually all residential broadband subscribers are under the impression that they really get the full use of the advertised bandwidth; it is a shock to most when they learn about oversubscription practices and QoS congestion management.
