As Benoit says, we were kind of taken by surprise. His post Is the Bandwidth Hog a Myth? and my Congestion Neutrality were picked up by Ars Technica, Slashdot, Gizmodo, Techdirt, DSL Reports and are being discussed on various fora. Thanks to all who picked it up, whether they agreed with our assertions or not.
I will repeat it again, we are not affiliated with anyone or getting money from anywhere whatsoever.
My interest is intellectual: the subject of oversubscription and Congestion Neutrality is in my opinion something which cannot be ignored, given the fast increase of access line speeds and the changes in how we use the Net, particularly the rise of video and real-time-streaming.
As Benoit is digging more into the data he will be getting from ISP's (kudoos to the 2 who already have committed themselves to generate real-life data), I will be focusing on congestion management and congestion neutrality.
Reading through comments I see 2 main lines of discussion.
"TCP/IP is implemented on zillions of computers, you have to change all them to get another method of congestion management which is near impossible".
Let's turn the argument around. The underlying thesis of this argument is that any variation in congestion management or traffic shaping or rate limiting would disrupt TCP/IP based communications.
It is clear from all the existing examples ( like Comcast's traffic management, traffic shaping and prioritizing tools, application based throttling some ISP's have implemented, the different method of congestion management in HFC networks in shared channels and so on) that it is very well possible to implement a variety of strategies of traffic shaping/congestion management without disturbing the TCP/IP based communications. Be it on level 2 or level 3, that is not the issue.
So why wouldn't it be possible to design and implement a better strategy?
The question if this added complexity is worth it, compared with just increasing middle mile capacity, is one where opinions greatly differ.
But: as soon as more sophisticated traffic management and congestion management is introduced, we badly need transparancy and Congestion Neutrality rules.
" We ISP's need data cap's to survive, otherwise we run into financial problems (high costs) or complaining customers"
Apparently this is dependent on your location and the resources available. In the Netherlands there are virtually no data caps on fixed line subscriptions, but we are blessed with high penetration, high density, short distances and the AMS-IX. But I can understand the predicament of an FttH ISP in Australia who has to deal with long distance backhaul and a very high percentage of traffic originating from US/UK/European websites. Transit costs are sky high so he has to somehow manage these costs. There are few options available now other than data caps. (The HongKong example of high speed local traffic and a relatively reduced bandwidth to the rest of the Internet is one that you can find in more places.)
But data caps is very crude and does not adress the real issue. The real issue is that our use of the Net has changed so the nice statistical multiplexing rules do not apply any more for oversubscribed middle miles. So do you tell your customers that you can have this wonderful access speed provided they behave themselves as in the past, averaging only 2-3 % or less of the datarate? Do you apply a crude and brutal rule to punish them? Or do you innovate to deal with the issue?
I am sure these ISP's would jump at a technological solution that would be better suited to their real needs and would satisfy their customers changed demands and behaviour.
To define 2 extremes. If all users would only use their internet access one hour per day, but all at the same time, they would experience a lot of performance problems. (And the ISP would have to pay more if he has a contract with a middle mile provider that is rated at the 95 % peak of traffic.)
If several users would max out their access line every day from 24 hrs to 7 hrs, no one would notice it, no costs would be incurred.
I can imagine more intelligent ways of dealing with congestion, like making sure latency stays low, like prioritizing users who have not been very active in the near past if the bottleneck fills up.
And last but least: increase the capacity....Moore's law still applies, we get double the bandwidth for the same price every 18 months.
Leave a comment