This is just flawed. It will put your provider that tends to sell 3mbps DSL lines that are rock-solid 24/7 with great peering below that of providers that sell 50Mbps lines that have fairly crappy peering and you only get 4Mbps and 5% packet loss during peak.
Or at least it will look that way on these graphs, since very few HD streams will be performed by the first provider's customers.
I'd be a heck of a lot more interested in looking at data that shows buffer underruns based on time of day. That will show congested providers vs. providers that simply offer less bandwidth. AKA it will separate the providers that lie, and those that are honest.
I don't know much about youtube's architecture, but in general these stats are pretty simple to do via basic math and access logging. You look at the total request time, how many bytes were sent, video bitrate, and then calculate how many seconds of video were actually downloaded vs. how long the request took. If the request took longer than the total seconds of video downloaded, then the user probably experienced buffering. It's a bit rough, but the numbers are incredibly useful in aggregate. I've used this method to identify peering issues (looking at you FT) when more advanced methods are not available.
Or at least it will look that way on these graphs, since very few HD streams will be performed by the first provider's customers.
I'd be a heck of a lot more interested in looking at data that shows buffer underruns based on time of day. That will show congested providers vs. providers that simply offer less bandwidth. AKA it will separate the providers that lie, and those that are honest.
I don't know much about youtube's architecture, but in general these stats are pretty simple to do via basic math and access logging. You look at the total request time, how many bytes were sent, video bitrate, and then calculate how many seconds of video were actually downloaded vs. how long the request took. If the request took longer than the total seconds of video downloaded, then the user probably experienced buffering. It's a bit rough, but the numbers are incredibly useful in aggregate. I've used this method to identify peering issues (looking at you FT) when more advanced methods are not available.