While the unraveling of net neutrality in the US has sparked fierce debates the world over, there has been little in the way of solid scientific evidence to establish just how pervasive the practice is among ISPs. It turns out the tier 1 providers are pretty picky when it comes to throttling video streaming services and such ominous results will have left a few Republicans with Pai on their faces this week.
AT&T, Sprint and Verizon have all been caught red-handed throttling OTT video services over their networks. Of course, this finding itself is no surprise. But one of the most intriguing findings from a year-long study conducted by the Northeastern University and University of Massachusetts Amherst, shows that T-Mobile, the ever-disruptive US mobile uncarrier, was identified as one of the most prevalent throttlers of Netflix. The damning results have arrived conveniently soon after parent company Deutsche Telekom was ordered to remove the throttling element from its zero-rated StreamOn service.
So, while damning data shows that AT&T was found to throttle Netflix 70% of the time and YouTube 74% of time the time, T-Mobile was a unique case in the manner in which it throttled video streaming. Research identified T-Mobile carrying out delayed throttling of Netflix, whereby the first few seconds of the transfer included rates up to 20 Mbps, after which they dropped to 1.5 Mbps.
By tuning an algorithm called PELT (Power of the Pruned Exact Linear Time), researchers were able to detect change points from tests where Netflix streaming was replayed on T-Mobile’s network using lab devices, with most change points detected around 7 MB. The methodology determined whether the change points for the ISP-app pair were randomly distributed or instead clustered together around a time or number of bytes.
It notes that T-Mobile implements delayed throttling based on bytes and, interestingly, researchers found that not all throttled video streaming services are treated equally under this policy. It detected 7 MB of delayed throttling for Netflix and NBC Sports, and 6 MB for Amazon Prime. YouTube was not found to be subject to any delayed throttling in the dataset, instead found to be throttled from the start in every instance.
While packet loss was zero during the delayed throttling period, immediately afterward the retransmission rate was 26%, eventually reducing to 17%. By comparison, YouTube initially experienced a loss rate of 6.8% and dropped to 3% after 70 seconds. In both cases, losses wasted substantial bandwidth, but the problem was more acute for cases with delayed throttling due to TCP sending at a high rate and adapting slowly, according to the study.
It’s worth noting that researchers intentionally purchased prepaid plans with mention of indicators for throttling practices, such as “video streaming at 480p” – as with T-Mobile’s BingeOn – or “video-optimized streaming” as throttling is often otherwise described.
The study included two sets of tests for Vimeo (with two different domains) and Amazon Prime Video (one using HTTPS and one using HTTP) in January 2019 to reflect the change in how the service delivered video that month. For T-Mobile, results showed that Vimeo tests were throttled only after November 2018, while a small fraction of YouTube tests were throttled at 2 Mbps (instead of 1.5 Mbps) only in January 2019.
Verizon and T-Mobile were both found to have throttled Amazon HTTP and HTTPS, while AT&T only throttled Amazon HTTP. An older recording of Vimeo traffic was throttled only by T-Mobile throughout testing in January 2019, but the updated recording of Vimeo was not throttled by T-Mobile, or any other carrier studied for that matter, which it believes was due to the DPI device in T-Mobile not updating its rules to reflect Vimeo’s change.
Naturally, here at Faultline Online Reporter we have focused on US results from the report, but the wider research in fact identified differentiation in both cellular and WiFi networks from 30 ISPs across 7 countries and 126,000 smartphone users, using an app called Wehe.
Author of the study David Choffnes, an associate professor at Northeastern University, implored the importance of continuing to publish the work, “It would be nice if this is not completely forgotten. At least when there’s an appetite for legislation on this topic, we’ll have the data.”