-----Original Message----- From: Owen DeLong [mailto:owen@delong.com] Sent: Monday, September 20, 2010 10:43 AM To: William Herrin Cc: NANOG Subject: Re: Did Internet Founders Actually Anticipate Paid,
On Sep 20, 2010, at 8:59 AM, William Herrin wrote:
On Sat, Sep 18, 2010 at 2:51 PM, Tony Varriale <tvarriale@comcast.net> wrote:
Of course the high level of oversub is an issue....
We'll disagree then. Oversub makes access affordable.
Sure, at 10:1. At 100:1, oversub makes the service perform like crap. With QOS, it still performs like crap. The difference is that the popular stuff is modestly less crappy while all the not-as-popular stuff goes from crappy to non-functional.
Only if the QoS is tilted in favor of the popular stuff. The concern here isn't QoS in favor of the popular stuff... The concern here is QoS in favor of one particular brand of service X vs another. (e.g. Netflix vs. Hulu).
If QoS favors unpopular but more profitable services, it can make the user experience for those services significantly less crappy than the competing more popular services and actually drive shifts in consumer behavior towards the less popular services.
Of course, as this succeeds, it becomes self-defeating over the long run, but, only if your goal is to provide good service to your customers.
If your goal is to keep your customers spending $minimal per month and stay attached to your service while using QoS payments from content providers to drive much larger margins, then, you can make a circuit through the content providers watching each one's popularity wax and wane as you screw with their QoS based on the money you get.
This is very bad for the consumer and, IMHO, should not be allowed.
In my career I've encountered many QOS implementations. Only one of them did more good than harm: a college customer of mine had a T3's worth of demand but was only willing to pay for a pair of T1s. In other words, the *customer* intentionally chose to operate with a badly saturated pipe. QOS targetted only at peer to peer brought the rest of the uses back to a more or less tolerable level of performance.
You are still making the mistake of assuming that the ISP is interested primarily in providing good service to their customers. When you move this from customer-oriented good service model to profit-oriented model built around keeping the pain threshold just barely within the consumer's tolerance, it becomes an entirely different game.
Owen
Devil's Advocate here, What would you say to ISP A that provided similar speeds as ISP B, but B took payments from content providers and then provided the service for free? Gives you the choice, ISP A, which costs, and ISP B, which is free, and most people wouldn't know the difference. ~J