I'm surprised, I thought it was run by Netflix so that they'd also have to boost all Netflix traffic. Maybe they are doing DPI? Netflix also has a "test video" somewhere that is probably a more representative test.
To figure that out, you'd need to do a speed test to the target host, then do another one while simultaneously running Speedtest.
Your Internet speed is also affected by the peering agreements your ISP has, i.e. how "well connected" it is to the ISP of your target host. Speedtest targets (and Fast.com) are pretty well-connected; i.e. they have a lot of peers, so you can get to them in relatively few hops. Many times, they're even run by your ISP, so the number of hops is minimal.
I've noticed there's a big difference in both bandwidth and latency based on the network my requests actually go out on (from my university Internet connection). For some reason, the Univ. of Cal. Network > Commercial ISP > Internet2. (Not sure why Internet2 sucks so much, but it seems really slow for some reason.)
I've observed this effect as well. For example, on wifi at a conference hotel, connections to everything else would significantly improve for a while if I ran Fast.com's speed test in the background. I used this trick throughout the week and it was very reliable.
Wow. I didn't know this. I just tested my FIOS connection. I pay for 75/75.
I did an mLab test (not well-known, but Google uses them). It said my download was about 30MBpS.
Then I ran the OOkla Test (well-known). It said I was getting 75.
Then I ran the mLab again, and this time, it said I was getting 55.
True, but I'm not a scientist, I'm certainly not claiming that this was done to be scientific (it was done in five minutes, while reading an HN thread), and this isn't worth doing much more than that.
If I care enough, I know how to do it a lot more exhaustively. For example, I'd probably write my own speed test, and host it in a known place. Relying on known tests; even obscure ones, is not something that I'd suggest for someone interested in The Scientific Method.
The results of that Ookla test suggest that there may some "emissions testing" stuff going on.
If it really is every single time, sure. My point is just that from one data point you can't draw the conclusion that there's something hinky going on. To really determine this you'd need to apply some pretty stringent statistical tests because there is a lot of potential bias that goes into any casual observation of this phenomenon.
> A Comcast customer who is dissatisfied with Internet speeds set up a Raspberry Pi to automatically tweet at Comcast each time speeds are much lower than advertised.
> "I pay for 150Mbps down and 10Mbps up," Reddit user AlekseyP wrote over the weekend. "The Raspberry Pi runs a series of speed tests every hour and stores the data. Whenever the down[load] speed is below 50Mbps the Pi uses a Twitter API to send an automatic tweet to Comcast listing the speeds. I know some people might say I should not be complaining about 50Mbps down, but when they advertise 150 and I get 10-30 I am unsatisfied."
I was doing this for a very long time. I had a speedtest run on my NAS, saved the results to a csv, and then tweet them at verizon. a few others have written similar scripts, it’s very interesting having that historical data, and Verizon reps on twitter actually tried talking to the bot.
I have always thought this conspiracy. I have Comcast Gigabit and more often than not the first speed test always caps around 250-275. If I run more than 2 times I get 980-990 on average. I wondered if it was some kind of prioritization at play here.
I've had the same experience though I only noticed that the speed test and pages related to it ran very fast but I got no comparable boost to other traffic.
