Monday, 18 August 2008

Say no to streaming video (there is no use for it and it hurts net neutrality)

Andrew Odlyzko has written a very insightful paper called "The delusion of net neutrality". The title is a bit of a misnomer as he is not arguing that net neutrality is a delusion. Instead he picks on network providers who argue that there shouldn't be/cannot be net neutrality. The technologies used for Deep Packet Inspection are used to enable streaming and real-time communications. Real-time communications may be sensitive to jitter, latency etc. and may benefit from QoS mechanisms on an overloaded link, but any communication that is not realtime (most streaming and near-realtime) communication shouldn't have that problem. Andrew Odlyzko'sargument is very true and well worded.

Video shouldn't be streamed it should be send via "faster than real time"-download. If it is send in this way their will always be way more in the buffer, so that temporary hickups will not harm the download and once the entire movie or a section of a movie has been downloaded there is no problem at all anymore even if the network goes down. The advocates of streaming video can't bring anything against this idea, except that it doesn't allow them to sell very expensive streaming kit. A big online video farm like Youtube always makes use of faster than real time downloads for exactly those reasons. Highest Quality of Experience for the end-user and lower costs for Youtube are a killer combination. So needing to enable streaming video is not a good argument in favor of non-net neutral networks.

One argument I did miss is that "Faster than realtime" also frees up resources later on in time, allowing better overall quality of experience to end-users. I've argued this in the OECD report on Broadband I wrote. Let's say a streaming movie takes 10% of line capacity. If there are 10 parallel streaming movies at the same time, the link is full. If you allow for faster than real time downloads, if there are two users they can download the movie to the buffer five times faster than they need it from buffer. When later on in time more people come online to watch a movie, the earlier downloaders either may have finished the download already, or may be able to stop the download of the movie for a time, freeing up resources to allow an eleventh or twelveth downloade in at the same time. (for other arguments, see my rants on QoS and flow based routing)