3 ms·

That's not what's expected. What's expected is that you get what you pay for, whether you're a customer on the hosting end or the residential end. If you're pay

That's not what's expected. What's expected is that you get what you pay for, whether you're a customer on the hosting end or the residential end. If you're paying your data center for 1Gbps, then they ought to have at least 1Gbps capacity to the major residential ISPs. If you're paying your residential ISP for 15Mbps, then they ought to have at least 15Mbps capacity to their peers. If there are 100 people trying to use 15Mbps at once at peak hours, they need at least 1500Mbps of capacity to provide the level of service they sold. What Comcast is being chastised for, the thing that forced Netflix to transit with them directly, is being paid by its customers to connect to the internet, then refusing to use that money to buy a pipe large enough to deliver that. It wasn't Netflix that wasn't buying a large enough pipe to send its own bits. If Comcast had resolved its disputes with Cogent and widened the pipe to meet the customer demand it's being paid by those customers to meet, I'm sure Netflix would've been happy to stick with its ISP.

sbd 13y ago
Unfortunately you don't buy an always open 15 Mbps pipe, you buy "15 Mbps peak speeds", or at least that's how it is worded for me. Which means at peak usage times you expect to get lower than 15 Mbps.
- dangrossman 13y ago
  It's hard for customers to buy that when the YouTube and Netflix videos they're watching only need a few hundred Kbps, but they can't get one good stream going on an otherwise unutilized connection. It's not a capacity crunch near the edges of Comcast's network, where lines are still shared in neighborhoods and upgrades cost the most; it's just greed, fattening already healthy profit margins by refusing to increase peering capacity and demanding peers pay more for bits customers are already paying for.
- Aloha 13y ago
  Which means deciding what an acceptable QoS is for those peak times - many of us think they decided too low - I personally think you should be able to get between 1/3 and 2/3rds (ideally around half) of 'peak' speeds on a sustained level at any time for any amount of time. It's just like the telephone network of yore, no single trunk was ever capacitied to mothers day call traffic - there were reroutes, and more than one path to any large point, so you'd get a lower quality of service, but the traffic would still get thru.
- betterunix 13y ago
  It is worse than that. "15 Mbps peak speeds" is actually "15 Mbps bursts." I am fine with getting less than my maximum speed; I understand that the connection is shared and that my neighbors' service can interfere with mine. TCP can deal with congestion. What grinds my gears is that ISPs are not content to sell me a maximum speed; they want to oversubscribe, then claim that they are selling burst service, then call it "unlimited super-awesome Internet!! 15Mbps!!" then penalize customers who fail to limit their use of the supposedly unlimited service.
zobzu 13y ago
tldr: peering issue, right? google has similar issues with some EU providers in particular in france, for youtube. traditionally the ISP pays to upgrade the pipes for the content that is provided by a third party (ie ISP pays for youtube, netflix traffic). ie the one whos downloading pays for upgrading pipes. thats where some ISPs (comcast too then i guess?) generally cringe: this way of dividing costs was ok 10 years ago, but not anymore: streaming is 90% of their traffic and its not sustainable.
- falsecelaona 13y ago
  In most cases, streaming is basically no different than downloading, right? The vast majority of people watch a video only once. So it's more like saying "downloading is 90% of their traffic and its not sustainable."
  - dannypgh 13y ago
    Sure, but it doesn't have to be that way. Something built around multicast groups could be used to stream multiple people the same content in a drastically more efficient manner, and then it could be stored locally on clients for time-shifting purposes. If you're trying to optimize for efficient use of limited bandwidth, unicast transferring of identical content to many many people is pretty wasteful. I think Netflix would argue that network links should be getting bigger and fatter to render the point moot, but given the streams are also getting bigger (we didn't always stream 720p everywhere, did we?) that would certainly take a lot more investment than what's happening now.
- waps 13y ago
  It's worse than that. If you look at how much bandwidth is required for any-to-any connectivity you can approximate how much it costs by modelling the internet as a full graph. Of course the cost of a full graph internet is n(n-1), with n the number of customers. The problem is, what happens when you connect a new customer. Well, simple : 1) the income of the ISP increases by some fixed amount 2) the cost of providing this connection increases by n+1 (the number of existing customers) So here's the catch np needs to pay for n(n-1)c pipes Where n is the number of customers on the internet. P is the average income per customer and c is the cost of a pipe that will sustain one customer. Needless to say, this won't work.
dingaling 13y ago
> If there are 100 people trying to use 15Mbps at once at peak hours, they need at least 1500Mbps of capacity to provide the level of service they sold. No retail ISP on this planet offers 1:1 ratio like that to its general subscribers. You can certainly buy uncontended service, but it will start at several hundred dollars per month for about 2 Mbps.

sbd 13y ago

Unfortunately you don't buy an always open 15 Mbps pipe, you buy "15 Mbps peak speeds", or at least that's how it is worded for me. Which means at peak usage times you expect to get lower than 15 Mbps.

dangrossman 13y ago
It's hard for customers to buy that when the YouTube and Netflix videos they're watching only need a few hundred Kbps, but they can't get one good stream going on an otherwise unutilized connection. It's not a capacity crunch near the edges of Comcast's network, where lines are still shared in neighborhoods and upgrades cost the most; it's just greed, fattening already healthy profit margins by refusing to increase peering capacity and demanding peers pay more for bits customers are already paying for.
Aloha 13y ago
Which means deciding what an acceptable QoS is for those peak times - many of us think they decided too low - I personally think you should be able to get between 1/3 and 2/3rds (ideally around half) of 'peak' speeds on a sustained level at any time for any amount of time. It's just like the telephone network of yore, no single trunk was ever capacitied to mothers day call traffic - there were reroutes, and more than one path to any large point, so you'd get a lower quality of service, but the traffic would still get thru.
betterunix 13y ago
It is worse than that. "15 Mbps peak speeds" is actually "15 Mbps bursts." I am fine with getting less than my maximum speed; I understand that the connection is shared and that my neighbors' service can interfere with mine. TCP can deal with congestion. What grinds my gears is that ISPs are not content to sell me a maximum speed; they want to oversubscribe, then claim that they are selling burst service, then call it "unlimited super-awesome Internet!! 15Mbps!!" then penalize customers who fail to limit their use of the supposedly unlimited service.

zobzu 13y ago

tldr: peering issue, right? google has similar issues with some EU providers in particular in france, for youtube. traditionally the ISP pays to upgrade the pipes for the content that is provided by a third party (ie ISP pays for youtube, netflix traffic). ie the one whos downloading pays for upgrading pipes. thats where some ISPs (comcast too then i guess?) generally cringe: this way of dividing costs was ok 10 years ago, but not anymore: streaming is 90% of their traffic and its not sustainable.

falsecelaona 13y ago
In most cases, streaming is basically no different than downloading, right? The vast majority of people watch a video only once. So it's more like saying "downloading is 90% of their traffic and its not sustainable."
- dannypgh 13y ago
  Sure, but it doesn't have to be that way. Something built around multicast groups could be used to stream multiple people the same content in a drastically more efficient manner, and then it could be stored locally on clients for time-shifting purposes. If you're trying to optimize for efficient use of limited bandwidth, unicast transferring of identical content to many many people is pretty wasteful. I think Netflix would argue that network links should be getting bigger and fatter to render the point moot, but given the streams are also getting bigger (we didn't always stream 720p everywhere, did we?) that would certainly take a lot more investment than what's happening now.
waps 13y ago
It's worse than that. If you look at how much bandwidth is required for any-to-any connectivity you can approximate how much it costs by modelling the internet as a full graph. Of course the cost of a full graph internet is n(n-1), with n the number of customers. The problem is, what happens when you connect a new customer. Well, simple : 1) the income of the ISP increases by some fixed amount 2) the cost of providing this connection increases by n+1 (the number of existing customers) So here's the catch np needs to pay for n(n-1)c pipes Where n is the number of customers on the internet. P is the average income per customer and c is the cost of a pipe that will sustain one customer. Needless to say, this won't work.

dingaling 13y ago

> If there are 100 people trying to use 15Mbps at once at peak hours, they need at least 1500Mbps of capacity to provide the level of service they sold. No retail ISP on this planet offers 1:1 ratio like that to its general subscribers. You can certainly buy uncontended service, but it will start at several hundred dollars per month for about 2 Mbps.