Exploring Verizon Wireless’s new UDP Video Throttling – Net Neutrality concern?

On August 22nd, Verizon Wireless, the USA’s largest and most popular mobile carrier, changed around their smartphone plans. In a questionably competitive move, Verizon revamped their recently launched Unlimited Data plans to include Video Throttling (otherwise known as Traffic Shaping targeted towards video). This is similar to moves carried out by AT&T Mobility, T-Mobile USA, and Sprint in the past, with T-Mobile being the major proponent towards limiting video bandwidth. Verizon’s competitive advantage over the other three, up until recently, was their unthrottled stance towards delivering Internet content regardless of what the data is for – video streaming, VPN, web browsing, gaming, or business data.

As a grandfathered unlimited data plan holder, I’ve never had the benefits of hotspot support (without tether hacks / rooting), but I’ve always held a mostly untouched, and an unfettered mobile data connection. One which, if I have to count on it to be there for me when I’m traveling or stuck somewhere far from home, I can use it as desired from my mobile device. It’s a nice convenience, and isn’t something I abuse – typically my heaviest usage is after 2AM, when most are in bed and the network is otherwise idle. The recent changes effectively remove the grandfathered unlimited data plan’s benefits, and Verizon has moved all functionality to match the new “Verizon Plan GoBeyond Unlimited” tier. Under the GoBeyond Unlimited plan, Verizon promises both unlimited 4G LTE access, as well as access to HD Video. The lesser plan, the “GoUnlimited” plan, features the same 4G LTE Access but access to SD Video instead.

Looking further at the details, Verizon has mention the following:

• Video will be limited to 720p for Smartphones lines, and 1080p for Tablets lines.
• Mobile hotspot lines and data rates for video will be limited to 720p quality.
• Devices on a limited data plan, are restricted to the same 720p video quality as an unlimited line is restricted to.
• Video quality above 1080p is effectively no longer happening. That includes 4K.
• Verizon will limited any suspected video streaming to a maximum of 10Mbps, if not initially flagged by their network.

With that said, this new move from Verizon runs foul of some past rulings about net neutrality, and regulations placed upon Verizon regarding their B13/700c LTE spectrum. But how far has Verizon gone in restricting video, and what services are affected? How is Verizon possibly detecting, and shaping traffic? This is what I’ve sought to explore.

Full disclaimer, I have some information gathered, however I have not completed any scientific verification of the data. This is a “first impressions” of Verizon’s new traffic shaping policies.

Getting a baseline, and beginning the tests

To first guarantee some accurate results, I made my way to a city area near our local University (SUNY Buffalo North Campus), which historically, has plenty of mobile network capacity to handle video streaming beyond 1080p reliably. I made sure I was within range of at least two mobile towers with ample capacity, and had a strong SNR (Signal to Noise ratio) to ensure my mobile data connection is not affected by poor signal quality.

To also provide some comparison between Hotspot vs. Mobile, I came equipped with a Samsung Galaxy Note 4, which is dated, but otherwise an acceptable baseline for the vast majority of phones out in the field today. The Note 4 has a formerly Grandfathered Unlimited data plan, with no Hotspot feature enabled. I also used a Netgear AC791L AirCard, which is equipped with it’s own 5GB mobile data plan, and a NuVision TM800W560L Windows 10 Tablet to represent a PC attached to the Verizon network. Note that the Windows tablet only has an 802.11n Wireless card in it, but it is sufficient enough to achieve speeds over 10Mbps.

The next step was to ensure the connection to the network is fast enough to determine whether or not Verizon is performing any traffic shaping. That means speed tests. Performing one test at a time. First, is Verizon’s own speed test, followed by testmy.net, and finally, a Google/MLab Speed Test.

Next, I start to check over a few services using the Verizon Wireless connection, since the connection is confirmed to be fast enough in the location. My expected result is that all video services, no matter where I go, should hit “720p” or about 4Mbps, according to Verizon’s marketing. I also assume, given how Verizon’s verbiage of detecting video streams exists, that traffic which looks like a video stream should not exceed 10Mbps.

My first test begins with Netflix’s Fast.com speed test. While I am not a Netflix customer, the fast.com speed test operates off of the same video servers Netflix uses to deliver video. This will help identify whether Verizon is shaping based on IP address, or based on the content of the data by means of Deep Packet Inspection. As a control, I run the test twice.

Practically the same result, given margin of error. As seen in the Network Activity, the bandwidth spikes only briefly above 4Mbps, but otherwise, the data is only allowed to flow at a rate of 4Mbps. This is about enough to support a decent quality 720p, 30FPS, non-HDR video stream, which supports Verizon’s claim, had I been playing Netflix.

The next test is to attempt to play a YouTube video from the same connection, now that the traffic shaping has been confirmed to be occurring with Netflix.

Curiously, this video was able to play at 1080p, 30FPS without having to stop and buffer up additional video content. This is likely because the Linus Tech Tips video being showcased, is both well optimized for bandwidth vs. quality by avoiding excessive motion, but also due to the codec being used (Google VP9) being capable of better compression efficiency. As noted by the network activity graph, the 4Mbps video shaping is also applied to YouTube content. The spikes seen in the graph are likely due to how YouTube serves up chunks of the video in each HTTP request, thus each request corresponds to a spike. Additionally, because of Wi-Fi and Cellular data networks with forward error checking, these spikes could simply be due to pauses in connectivity followed by a burst of data from a buffer.

Despite 1080p, 30FPS playback, I then attempted to try 1440p, 30FPS content (not to worry, the Tablet can handle it – Chrome isn’t good at GPU Decode offload). The video did end up buffering, even with the VP9 codec’s efficiency, and with a steady 4Mbps of data flowing in.

Next, I moved onto another video streaming service which is popular with the gaming community, called Twitch. Twitch works similarly to YouTube or Netflix, where pre-recorded videos are available for streaming, but more popularly, it is used for live streaming of eSports gaming events. Gamers, such as I, love to watch on Source quality, not only because it tends to work better with the site (Twitch historically has trouble re-encoding streams to lower quality), but because the video quality is that much better – 1080p, 60FPS, depending on the stream. This weekend, the League of Legends North American Qualifiers are on. So why not use that, a very popular stream, as a baseline?

Surprisingly, it plays. It plays flawlessly. It’s playing with data flowing in at rates above 10Mbps. There is absolutely no sign of traffic shaping like there was with Netflix or YouTube. YouTube, Netflix, and Twitch all deliver video over HTTPS / TCP streams. So why is Twitch able to play without traffic shaping? Is it possible Verizon is being choosy as to what they shape? Is it merely a limitation of their shaping implementation, where “Detection” is just fluff, and they are instead applying the shaper to data from certain IP Addresses? Are they preferring certain services? Let’s try that again with another Twitch stream.

Although I went down the totem pole in terms of quality, the same symptoms are seen. There is no traffic shaping applied to Twitch at all. It’s playing just fine, and data is bursting in at above 10Mbps regularly, and often. Since Twitch seems unaffected, why not try some services which are owned by Verizon, to see if their services are being applied to the throttle? We’ll use AOL and Yahoo services, since both companies are owned by Verizon.

The final results here are mixed. It’s apparent that videos served from Yahoo are not limited to a 4Mbps rate as otherwise seen with YouTube and Netflix. On Yahoo news, I can exceed 10Mbps and remain there while buffering up the video. Although my screenshot doesn’t show the video playing, please be assured that it was in fact, playing without an issue. My screen capture utility did not capture the video playback. On the Flickr side of things, the throughput seems to be struggling, but the rate is otherwise above the 4Mbps limit seen on YouTube and Netflix. Finally, on the AOL Side of things, the performance is even worse, but there is no immediate sign of shaping based on the network activity – but this could be due to how AOL Serves video, in a more bursty, but controlled manner. These results were seen consistently across other videos hosted by both sites.

Returning to Netflix, YouTube, and Twitch performance, given the mixed results with our other services, I then switch to checking services from my phone. Here, I’m totally expecting to see everything limited to 720p regardless of the source. I run through many of the same tests again from the phone.

With all said and done, the results are practically the same from the Smartphone, as they were on the Windows tablet tethered to the Netgear Hotspot. Netflix and YouTube are capped off at around 4Mbps (this includes YouTube Gaming, as that is served from the same location as traditional YouTube). Twitch and other services like Speed tests, are left untouched. 1080p video or higher on the phone, is a big mixed bag as to whether it will play or not.

Conclusion

With my quick and non-scientific checks done, it’s pretty safe to say that most of Verizon’s claims are true as of right now. To outline it all:

• Video is confirmed to be limited to 720p by means of traffic shaping. This allows for 720p playback, and sometimes 1080p playback based on the codec and video itself.
• Verizon does not re-encode the video. As video is delivered over HTTPS, it is much more expensive, and difficult to limit video through re-encoding. In the past, Verizon used to re-encode video sent to mobile phones via a Man-in-the-middle Proxy, which still exists on their network to this day.
• A 5GB Data limited Hotspot, and a Grandfathered unlimited smartphone line, behave the same.
• It appears that only mainstream streaming services like YouTube and Netflix have a 4Mbps limit applied.
• Services operated by Yahoo and AOL are not affected. Twitch is also unaffected.
• The Netflix speed test is affected, while other speed tests are not. This is due to Netflix’s speed test being hosted off of the same servers used to stream video to Netflix customers.
• The traffic shaping is not triggered by directing customers through specific DNS resolvers – Earlier in my tests I had changed the Hotspot to only resolve queries against OpenDNS.

With that said, there’s a few issues here. Verizon appears to be limiting the video based on it’s source, meaning IP address. There is likely a blacklist and whitelist being maintained by Verizon, and updating this list is less about automation, but about whether or not Verizon customers are using a service heavily enough to be deemed “impactful” to the performance of the network.

This, either way, doesn’t set a good precedence for the future of the wireless network. Although competitors were beginning to limit video playback based on capping off data performance, Verizon was the remaining carrier (with arguably the oldest LTE network out there) without any sort of traffic shaping implemented. Given how choosy Verizon currently is with what is slowed and what isn’t, this is likely going to pave the way for partner data, which is given a higher bandwidth allocation (Verizon is already doing something similar with a program they call FreeBee) to those who pay for quicker delivery. Verizon’s implementation appears to be incomplete, and going forward I expect these results to change. Perhaps, though, the pick and choose game seen here has been chosen to avoid backlash from a specific group of customers which could otherwise tarnish the brand, like Twitch viewers.

This behavior is concerning. Verizon is currently bound to an FCC license around their 700c LTE block which indicates that Verizon cannot throttle an application provided it is being used on a legal basis, and all services of a specific application must be treated equally. This is currently not the case, and there’s a possibility of a future court case around this from groups such as the EFF.  Although given the current status of the FCC here in the United States, it’s unlikely the FCC will proactively do anything… if at all. Mobile networks are not Title II Classified, and thus are not necessarily covered under previous/existing Net Neutrality rules, which is how the industry has been able to get away with this behavior. Verizon can technically get away with this by claiming their traffic shaping of video is a form of network management, by providing suitable levels of service in order to control bandwidth consumption, but again, they are currently running foul of net neutral behavior by choosing only specific services. Many folks could likely make the case that Verizon needs to focus on their infrastructure more instead of taking moves like this. More aggressive upgrades, less over-engineering of the network, and more honest sales pitches.

As a solution, Verizon should consider selling mobile data based on rated speed, instead of based on the Gigabyte. Similar to home data connections. Shaping based on the device type or plan would satisfy the net neutrality issue this causes, and would help provide better network management with less administrative overhead. Shaping should be done to maximize a device’s capabilities (for example, phones with 1440p displays should be able to play 1440p). Additionally, any traffic shaping should not apply when traffic levels are low. Such as during the even hours, where the network in many areas is idle. This is a concept which has worked for many ISPs on the wireless side, and has also worked for the Satellite guys to allow heavier customers to offload their heaviest data usage to the night time hours.

My remarks

On the note of Verizon’s claim regarding whether or not 720p vs 1080p is noticeable on a phone, I ask that someone with good eyesight take a look at content from something other than an iPhone. Even with modern iPhones, the display resolution is at best, 1080p, and the displays, despite being of good quality, are still IPS. The difference between 720p and 1080p can be readily seen when comparing from a device sporting an AMOLED display, and a screen at 1080p or above, like mine. In addition, on many services, there are differences in audio quality and color accuracy. Streaming apps will choose the best resolution based not only on the bitrate which is able to be sustained, but also the screen resolution. Which on an iPhone… means 720p unless an iPhone 7+ is handy and the app permits.