Comcast to 'Manage People, Not Protocols' > Atta Boy, Tony--

said by hobgoblin:

The more p2p uses move to DSL and Fiber the better for cable platforms.

said by »blog.wired.com/27bstroke6/2008/0···ity.html :

In response to a Charter Communications acknowledgment last week that it planned to let a third-party behavioral tracking company analyze its users' traffic for advertising purposes, Reed said the model may make sense to cable companies that are accustomed to inserting, and even overwriting, ads in television shows.

Good....Losing a couple of % of their customer base to free up 60% of their bandwidth makes perfect sense.

said by funchords:

Especially since no reasonable facsimile of the bandwidth they were sold was never intended to be delivered to them in the first place.

said by funchords:

In that video, he aptly explains the focus the company is putting on making sure that the United States continues to fall behind the rest of the world in speeds and access.

said by funchords:

Comparing my 6 Mbps/384 Kbps to Japan's 100 Mbps access? What a crock of crap!

said by funchords:

And, by the way -- yes, Japan says that a good fraction of their bandwidth is consumed by P2P. But they also ARE NOT throttling it.

said by espaeth:

Interesting spin on the video, but you missed the underlying theme. "Managing capacity" does not mean "stop infrastructure improvements".

said by funchords:

The only way out of congestion is building out of it.

said by funchords:

"Managing Capacity" is just shuffling the deck chairs on the Titanic -- in the end, the size of the demand is still the same, you're just forcing it through the bottleneck in a different order (or worse, dropping packets for sustained durations at a time).

said by funchords:

That's not management, that's mismanagement. Even worse, it's malmanagement -- since in the end it's the customer getting SCREWED.

said by espaeth:

I don't get you Robb. One minute you're arguing that providers should implement QoS, the next minute any network management is pointless. You argue that you want 100mbps broadband, but yet you want lower oversubscription to the edge. These are all diametric opposites when the constraints of reality are applied.

said by funchords:

They should implement QoS, but they should do it for the right reasons. QoS is about making services more robust during periods of congestion -- those rare moments when the network unfortunately starts queuing or dropping packets. These moments are bound to happen, since guessing capacity needs is an inexact science and network applications usually take as much as they can get anyway.

said by funchords:

But, when a network is queuing and dropping packets with increasing frequency and regularity, that's a sign that demand has reached supply. QoS doesn't fix this, and it never will.

said by funchords:

The non-"average" subscriber has rights, too. He has the right to access the bandwidth that he pre-paid for!

said by espaeth:

Most people get stuck in rush hour traffic at least twice a day, 5 days a week. Despite that being the case, you don't see them adding additional lanes constantly to work out the congestion. As long as the demand isn't constantly exceeding supply, eventually there will be ways to spread the demand out over time. Even highways tend to have types of QoS though -- motorcycles in CA are allowed to cut between cars in rush hour, busses in MN are allowed to drive on the shoulder to get around congestion, and car pool lanes are provided in many states to give people incentives to use efficient travel methods.

The situation is no different here. P2P may be a great application for sharing data, but the efficiency of distribution is about as ridiculous as cars that get 8mpg. Bandwidth to the edge is expensive to upgrade, and has limits on how far it can be expanded. Pushing the load to the edge is a design that lacks vision, and has presented scaling issues since it first started gaining popularity almost a decade ago.

What did he pay for? Speeds up to # mbps. If someone has been running their connection hot for the last 6 hours, why shouldn't they be scaled back a bit to allow the 30-minutes-a-day user to get decent speeds for the time they want to use the network?

said by funchords:

First off, it's all "edge." There is nothing at the middle of the internet.

said by funchords:

Secondly, you're sure discounting a lot of people who see that taking an archive and busting it into thousands of chunks in order to work very well in a congested network IS THE VISION that makes P2P viable competition for client-server.

said by funchords:

Next time your favorite Linux distro updates, download a copy the next way in each way and compare. So far, P2P kicks client-server's butt the last several Fedora or Ubuntu distros I've downloaded.

said by funchords:

And I'm okay with "up to" as long as what really happens is that the speed is achieved the vast majority of the time. Otherwise, it's just marketing trickery.

said by espaeth:

That's a false argument for efficiency, because if I wanted to do things quickly I wouldn't grab the ISO filled with 3+GB of dead weight packages I don't need anyway. A lot of people use their package manager to upgrade between versions anyway, which negates the ISO acquisition argument as well.

said by espaeth:

This is one of those paradoxical cases where you can only provide those rates most of the time if you keep people from using them to that level most of the time. The express (12 items or less) lane doesn't work if you have people show up with 4 carts with 12 items in each cart checking them out as single transactions.

said by funchords:

Thanks for the lists -- I understand what you are saying about them, but you also just made the case for why P2P is good for Minneapolis ... just think of all the connections not having to cross the boundary gateway toward Chicago or other CDN locations thanks owing to P2P providing a more local copy (of at least some of the data).

said by espaeth:

Hardly. With a well managed network the average subscriber wins. If I've got 10 heavy P2P users sharing my connection with the upstream CMTS, having some mechanism in place so that their 40+ TCP sessions don't clobber my single HTTP TCP session is a Good Thing™