I mentioned a couple of posts back that since the last HTC firmware update I’d been getting speeds in excess of 90mbit speed test results on Sprint no less.
This lead to a lot of tests and absurdly varying speeds from different testing apps and pages.
Since I have VPNs located all over the world, corporate cable, and last night I had access to a hospital’s multi-gigabit backbone WiFi I did some more testing and concluded that for the most part most testing services don’t return the info I wanted at that time.
I wanted to know how fast I was going from my phone to the tower to a nearby network server.
Checking Sensorly it was going from my house to the tower across Sprint’s network traveling down to Atlanta and then moseying up to Digitalocean servers in NY.
Total round trip was about 2400 miles. Download speeds were 42 mbit. This was the same basic speed on Comcast, Sprint, and VPNs in the US I was playing with, although the VPNs had some overhead that slowed them more.
The bottleneck on Sensorly seemed to be distance and no matter what speed connection I threw at it I wasn’t getting much above 50. This included my 250 internet connection, the hospital’s multi gig backbone, etc.
Checking testmy.net and using closer servers and multiple I managed to achieve about 60 mbit. Once again on pretty much anything I threw at it.
Checking Ookla’s speedtest app, I got 107mbit as a top speed on Sprint, 180mbit top speed on Comcast, and varying speeds depending on where in the world I was popping out on a VPN
What’s up with that?
There are a lot of factors to speed when it comes to your ISP/carrier/etc. Let’s look at what I was doing with Ookla’s Speedtest.
My cell phone was talking to nearby LTE towers, data was entering into Sprint’s Nashville data centers, exiting in Nashville to Comcast’s lines as the Ookla speed test server was nearby on Comcast (7 miles) and traffic was going probably a total of 14 miles and five hops across internet that probably has a Quality of Service rule set to prioritize traffic there as it’s going to look good for everyone involved. Results: 107+mbit
With Sensorly I was talking to the LTE tower, it was dropping onto Sprint’s network, traveling down to Atlanta, heading up to NY, exiting to the general internet and making about fifteen total hops before it reached its destination. Results were 48mbit.
Testmy.net ended up basically 30mbit on a server, but with multiple servers being tested at the same time I could get speeds over 60mbit. The total to any one geographically located server I could get was in the 30mbit range, but combined I think I topped out at 71mbit at one point.
A whole lot of factors to consider
You to tower:
We’ll start with you to base station as the first. Your base station can be WiFi, LTE, 3G, or tin cans tied together with string and two toddlers who don’t understand they have to pull the cans apart. This is your first leap out into the net.
You might be able to hear the transmitter at the base station fine, but it might not be able to hear you because it’s got quite a bit of power behind it and you’re trying to make a 2000mA battery last 16 hours.
Tower to internet:
Just because you can talk to the tower at a bajiggabit per second doesn’t mean the tower has that level of capacity behind it. An example is my LTE tower – I had full signal, theoretically should have been getting 20+mbit the last two years, but it was backed up by an internet connection that was slow and essentially useless.
My speed tests anywhere at my house were 0.3mbit. Calling in would always get a Sprint op saying something vague like that the tower was in the red, or that there was a connection issue on the back end, or that I really should dump them and go with T-Mobile.
So, if your tower can’t send and receive data upstream toward the internet side, your perfect signal is useless. Internet doesn’t hop cell tower to cell tower, it goes cell tower to cables to fiber buried in the ground.
Internet to testing site:
While theoretically we now have Net Neutrality, this doesn’t mean that everything is the same everywhere for everyone, and it’s generally not malicious intent.
Internet backbone companies transmit data back and forth and form peering agreements with other backbone providers. If data is cheaper to send via X, it’s probably going to go to X regardless of if Y is faster. At least until such time as Y becomes cost competitive.
In the example above where my data went south before going north, that’s probably the least expensive route. It may be the fastest, but I think faster would involve jumping in Nashville from Sprint to Metro-E to a AT&T line that heads directly to NY. Probably costs more and more headache.
Remember back when we were having weird issues with DDOS attacks on Pocketables servers for really no reason? This stuff goes on across the internet at all times. One of the hops along your route to the testing location may also be the victim of 12,000 computers running a DDOS attack through it.
imagine a four way stop and a herd of buffalo preventing you from getting through at a normal pace. That’s essentially it.
Real storms, drunk people, utility failures, exploding trucks, and ants have also been known to degrade the internet significantly.
The testing server itself:
Some testing servers also have a max bandwidth they’re capable of testing. Back a few years ago when a lot of these were set up the idea that every cell phone on the planet was probably capable of 150mbit internet connection wasn’t a huge consideration in the business plans to develop apps and websites, purchase bandwidth. Your average phone did a megabit and you had 20 people testing at a given point and paid for it with advertising and maybe selling performance data to the ISPs.
Over the past few years we’ve seen diminishing returns on advertising, networks finally bringing the speeds we’d been promised five years ago, and the minimum household speed upping by let’s just say some imaginary number that makes you cross your arms, tap your feet, raise an eyebrow and go “ah hurmm”. Yes, exactly that well researched number.
And a lot more people are running speed tests to see if they’re getting what they’re paying for.
How long the test is:
If you’ve ever watched a speed test you’ll notice that they start low and ramp up usually. I got to wondering if there was a way to run one longer whether this would get a true result as near the end of all of these I see the numbers are still rising.
Running Speedtest and then immediately running speedtest again I went from 99mbit to 107. My guess is that the LTE tower had focused in on my needs or that the backbone network had worked out a path especially for me to candyland.
Most likely the packet size is being adjusted (and the tower fell in love with me,) it’s just getting in the groove. Unfortunately I haven’t worked at an ISP since 2000 so all I can say is BGP, ATM, packet size, and “yeah, 3mbit DSL is the future. What would you need more than 3mbit for?”
Similarly speeds reported faster on testmy.net with larger manual 250MB test data sets.
What time is it?
One other huge factor is how many minutes has it been since Game of Thrones had a new episode released on HBO Go or whether Frank Underwood has been seen recently.
Another is is it rush hour and you have 80,000 people stuck in traffic a mile from your house on their cell phones streaming Game of Thrones.
I mentioned I had VPNs, and I used them. Something to remember with VPNs is you’re doing a whole lot more than just internet traffic. You’re now going across the internet to the location that the ISP is physically in, and then working your way back to the speed test.
A tracert on your VPN may show 12 hops to the speed test, but your connection took eight just to get to the VPN. While the VPN is handling a lot of the processing and data wrangling, you’ve got a lot of extra hops and CPU time making the encryption doughnuts, and then eating the encryption doughnuts.
Results however, can be mixed. Such as if you VPN has an exit right next to the testing servers and the VPN company has extremely good connection and you’re short hops to the VPN.
For my purposes on these three tests I’ve mentioned I got the following on many many tests last night:
Ookla’s Speedtest provided me the highest number based on servers being really geographically close and sometimes in the same network.
Sensorly provided me what I consider to be my average speed out to any one given site in the world.
testmy.net’s multi server test gave me a combo of Sensorly and Speedtest and I consider it to have reported my overall abilities to internet adequately.
I’ll note that what I experienced does not appear to be the case with Pocketable’s reader Deslok as his appear to have my findings for Sensorly and Testmy.net reversed.
Alternately I could have no idea what I’m talking about so take this all with a grain of salt.