Comcast extends data caps

[leadeater]

28 minutes ago, Bombastinator said:

It’s about finding the top1% of users that produce the lions share of the traffic, and moving them on to a different group

95th percentile would actually more accurately find these users, so long as you don't give them enough peak bandwidth to do a Google and do multiple TBs in a day fully committing the entire bandwidth of your internet plan. Both data usage billing and 95th percentile will not correctly identify these users.

 

Edit:

Remember when it comes to residential connections no one customer has equal or greater bandwidth than interconnection links within the network so a sample resolution of 5 minutes or something similar isn't really a problem. These networks are over subscribed on purpose like you know because not everyone is going to be using a lot of bandwidth at the same time. A typical GPON port is split 32:1 with a link bandwidth of 2.488/1.244 Gbps, typical plans offered are 1000/500 so would only take 2 customers to saturate that link. The network interconnect from that node is far higher more likely 1 or multiple 10Gbps connections.

 

There could be 32 GPON ports in that node servicing 1024 residential customers, at the node with a single 10Gbps interconnection it would only take 10 customers to saturate that link if those 10 had 1Gbps connections and were actually using 1Gbps each. Residential networks are so oversubscribed it really does make bandwidth usage over a month highly ineffective to manage bandwidth.

 

Thing is most services we use will not use a lot of bandwidth and even the ones that do are rarely capable of actually achieving 1Gbps, I can from speed test servers but say downloading an Ubuntu ISO not even close.

[Bombastinator]

8 minutes ago, leadeater said:

95th percentile would actually more accurately find these users, so long as you don't give them enough peak bandwidth to do a Google and do multiple TBs in a day fully committing the entire bandwidth of your internet plan. Both data usage billing and 95th percentile will not correctly identify these users.

So what they have isn’t good enough.  It’s merely all they’ve got.  One way to kill the rationalization would be to find one that actually does work then and point out that they should be doing it instead of the thing that doesn’t work. 

[leadeater]

21 minutes ago, Bombastinator said:

So what they have isn’t good enough.  It’s merely all they’ve got.  One way to kill the rationalization would be to find one that actually does work then and point out that they should be doing it instead of the thing that doesn’t work. 

Well we already have one that works better, 95th percentile. If that's not good enough then 96th or 97th etc. Thing is like my edit above at the end points out achievable bandwidth on many services and what some services require (HD video) is actually very low. The simple truth is ISPs do not need to nor have a problem of bandwidth such is why when data caps got removed due to COVID absolutely nothing went wrong.

 

Thing is if ISPs want to claim there is network bandwidth problems within the network instead of just believing them they should actually prove it. Once they can prove it's a problem and where those problem points actually are we can then assess the validly of this and if the proposed solution i.e. data usage charges will actually effectively address the issue or not. If the problem points in the networks are solely at the internet exchanges and bandwidth between networks (which is largely where it actually is) then that is so many point in the network chain removed from individual residential connections that putting measures in place to discourage people from using the bandwidth they have paid for will be highly ineffective. The most effective resolution is to either pay for more bandwidth between the problem network or provision another link (and pay for that bandwidth).

 

Internet exchanges are actually not short on bandwidth and ports to utilize, it's just that a lot of the companies don't actually get along very well along with ISPs in general being unwilling to pay for more bandwidth. Netflix too for that matter btw, they also have Transit and Peer agreements in IX's.

[Bombastinator]

11 minutes ago, leadeater said:

Well we already have one that works better, 95th percentile. If that's not good enough then 96th or 97th etc. Thing is like my edit above at the end points out achievable bandwidth on many services and what some services require (HD video) is actually very low. The simple truth is ISPs do not need to nor have a problem of bandwidth such is why when data caps got removed due to COVID absolutely nothing went wrong.

 

Thing is if ISPs want to claim there is network bandwidth problems within the network instead of just believing them they should actually prove it. Once they can prove it's a problem and where those problem points actually are we can then assess the validly of this and if the proposed solution i.e. data usage charges will actually effectively address the issue or not. If the problem points in the networks are solely at the internet exchanges and bandwidth between networks (which is largely where it actually is) then that is so many point in the network chain removed from individual residential connections that putting measures in place to discourage people from using the bandwidth they have paid for will be highly ineffective. The most effective resolution is to either pay for more bandwidth between the problem network or provision another link (and pay for that bandwidth).

 

Internet exchanges are actually not short on bandwidth and ports to utilize, it's just that a lot of the companies don't actually get along very well along with ISPs in general being unwilling to pay for more bandwidth. Netflix too for that matter btw, they also have Transit and Peer agreements in IX's.

I thought you said 95th percentile doesn’t work either.  

[leadeater]

17 minutes ago, Bombastinator said:

I thought you said 95th percentile doesn’t work either.  

It doesn't work for a specific situation if the peak bandwidth is sufficiently high enough. So far the only documented case of abusing it is Google, who had sufficiently large enough bandwidth. I don't think there will be many ISPs handing out that kind of bandwidth to residential customers.

 

The only method that is foolproof is one where you pay for the complete full bandwidth of your connection, you don't want that, nobody wants that. Not even the ISPs as the cost would put customers off which isn't great for making money.

 

Thing is the more sample points you take the harder it is to cheat it, or you can put other detections in place like doing the same percentile calculation but doing it over 7 days or some other length of time along the 30 day sample points and see if there is a large discrepancy between all 8640 samples and the start/middle/last 2160 of them. Or something like that. That's how Google was caught and then confirmed by watching it in real time happen.

 

But as per the original point, data caps don't do what the ISPs say they are for and they are not needed. If they were needed for that reasoning they wouldn't be using data caps. The solution does not fit the problem description.