3 ms·

"The article is not about congestion control but rather marketing choices where one streaming service is favored over the other within the captured market of an

by timr 8y ago

"The article is not about congestion control but rather marketing choices where one streaming service is favored over the other within the captured market of an ISP customers." I wonder if that's true, or this is simply a result of the peering agreements that content providers have engaged in since forever. For those who don't know: if Netflix is colocated in the same datacenter(s) as CSPire, it's possible to "throw an ethernet cable over the cage wall" and make a direct connection to CSPire. This results in dramatically better bandwidth and latency, but requires significant up-front costs (it's also not possible for every provider to peer with every distributor, for logistical reasons). As much as I believe in the concept of net neutrality, I feel like the conversation misses many of the technical nuances of the way the internet actually works. Peering is one example: you can have an entirely "neutral" internet, but still observe dramatic differences between one provider and another, based on your network location. (Edit: now that I've read TFA, it's pretty clear that they're not talking about favoring providers, but rather, throttling entire classes of traffic. Which is a different discussion.)

belorn 8y ago
Could there be other causes than differentiation to explain the data collect? That would be an excellent questioning to ask the team at University of Massachusetts if they made a dissertation or research paper, since all the article says is that they used a established peer-reviewed technique that assumingly has gone through such questioning. It would be interesting to know the statistical probability that collocated servers would exclusively explain the data. > now that I've read TFA, it's pretty clear that they're not talking about favoring providers, but rather, throttling entire classes of traffic. I must strongly disagree there. What kind of different class of traffic would give throttled traffic to different forms of on-demand video stream services? NBCSports, Netflix, YouTube Premium and Amazon prime. I don't see it so please name the four different classes of traffic those sites represent that a congestion system would use as identifiers.
- timr 8y ago
  "I must strongly disagree there." The article makes it clear what they did, and what conclusions they drew: "Using a previously established, peer-reviewed technique, the team conducted more than half a million data traffic tests across 161 countries. From this data, the team found that internet service providers are “giving a fixed amount of bandwidth—typically something in the range of one and a half megabits per second to four megabits per second—to video traffic, but they don’t impose these limits on other network traffic.” Moreover, the data shown directly below that paragraph reveals little pattern in the provider/app throttling rates. The data shows mostly that video is being throttled across the board. Could there be other explanations for this? Absolutely. Peering is just one of many plausible explanations.
- ti_ranger 8y ago
  > What kind of different class of traffic would give throttled traffic to different forms of on-demand video stream services? NBCSports, Netflix, YouTube Premium and Amazon prime. There may be a small amount of packet loss at the time due to the ISP shaping their traffic to prevent congestion. The higher the latency, the more the packet loss will affect the throughput. The services you mention all use different CDNs, which may all have different latencies at the point of observation: > NBCSports, Netflix, YouTube Premium and Amazon prime NBCSports - I would guess Akamai, who deploys at IX and ISP Netflix - Usually Openconnect CDN devices at IX or ISP YouTube Premium - Google Global Cache deployed at IX and/or ISP (GCCs are deployed in most consumer ISPs, probably the best-distributed CDN at the moment). Amazon prime - Cloudfront, deployed mainly at IX, I am not sure if they have deployments in ISPs yet. (I haven't tried to find more information on their approach, but https://dd.meddle.mobi/index.html https://dd.meddle.mobi/index.html seems to indicate that CDN issues shouldn't affect the way they measure, but until I do some checking on whether popular DPI platforms would be able to be fooled by this, and without them providing timestamped data, I'll have to test myself against DPI policies I wrote and I hope my ex-colleagues haven't changed too much ...)

by timr 8y ago

belorn 8y ago

Could there be other causes than differentiation to explain the data collect? That would be an excellent questioning to ask the team at University of Massachusetts if they made a dissertation or research paper, since all the article says is that they used a established peer-reviewed technique that assumingly has gone through such questioning. It would be interesting to know the statistical probability that collocated servers would exclusively explain the data. > now that I've read TFA, it's pretty clear that they're not talking about favoring providers, but rather, throttling entire classes of traffic. I must strongly disagree there. What kind of different class of traffic would give throttled traffic to different forms of on-demand video stream services? NBCSports, Netflix, YouTube Premium and Amazon prime. I don't see it so please name the four different classes of traffic those sites represent that a congestion system would use as identifiers.

timr 8y ago
"I must strongly disagree there." The article makes it clear what they did, and what conclusions they drew: "Using a previously established, peer-reviewed technique, the team conducted more than half a million data traffic tests across 161 countries. From this data, the team found that internet service providers are “giving a fixed amount of bandwidth—typically something in the range of one and a half megabits per second to four megabits per second—to video traffic, but they don’t impose these limits on other network traffic.” Moreover, the data shown directly below that paragraph reveals little pattern in the provider/app throttling rates. The data shows mostly that video is being throttled across the board. Could there be other explanations for this? Absolutely. Peering is just one of many plausible explanations.
ti_ranger 8y ago
> What kind of different class of traffic would give throttled traffic to different forms of on-demand video stream services? NBCSports, Netflix, YouTube Premium and Amazon prime. There may be a small amount of packet loss at the time due to the ISP shaping their traffic to prevent congestion. The higher the latency, the more the packet loss will affect the throughput. The services you mention all use different CDNs, which may all have different latencies at the point of observation: > NBCSports, Netflix, YouTube Premium and Amazon prime NBCSports - I would guess Akamai, who deploys at IX and ISP Netflix - Usually Openconnect CDN devices at IX or ISP YouTube Premium - Google Global Cache deployed at IX and/or ISP (GCCs are deployed in most consumer ISPs, probably the best-distributed CDN at the moment). Amazon prime - Cloudfront, deployed mainly at IX, I am not sure if they have deployments in ISPs yet. (I haven't tried to find more information on their approach, but https://dd.meddle.mobi/index.html https://dd.meddle.mobi/index.html seems to indicate that CDN issues shouldn't affect the way they measure, but until I do some checking on whether popular DPI platforms would be able to be fooled by this, and without them providing timestamped data, I'll have to test myself against DPI policies I wrote and I hope my ex-colleagues haven't changed too much ...)