Measuring Mobile Broadband Speeds

I was using Google search on my cellphone a few days ago and I thought my connect time was sluggish. That prompted me to take a look at the download speeds on cellular networks, something I haven’t checked in a while.

There are two different companies that track and report on mobile data speeds, and the two companies report significantly different results. First is Ookla, which offers a speed test for all kinds of web connections. Their latest US speed test results represent cellphone users who took their speed test in the first half of this year. Ookla reports that US cellular download speeds have increased 19% over the last year and are now at an average of 22.69 Mbps. They report that the average upload speeds are 8.51 Mbps, an improvement of 4% over the last year. Ookla also found that rural mobile broadband speeds are 20.9% slower at urban speeds and are at an average of 17.93 Mbps.

The other company tracking mobile broadband speeds reports a different result. Akamai reports that the average cellular download speed for the whole US was 10.7 Mbps for the first quarter of 2017, less than half of the result shown by Ookla.

This is the kind of difference that can have you scratching your head. But the difference is significant since cellular companies widely brag about the higher Ookla numbers, and these are the numbers that end up being shown to regulators and policy makers.

So what are the differences between the two numbers? The Ookla numbers are the results of cellphone users who voluntarily take their speed test. The latest published numbers represent tests from 3 million cellular devices (smartphones and tablets) worldwide. The Akamai results are calculated in a totally different way. Akamai has monitoring equipment at a big percentage of the world’s internet POPs and they measure the actual achieved speeds of all web traffic that comes through these POPs. They measure the broadband being used on all of the actual connections they can see (which in the US is most of them).

So why would these results be so different and what are the actual mobile broadband speeds in the US? The Ookla results are from speed tests, which last less than a minute. So Ookla speed test measures the potential speed that a user could theoretically achieve on the web. It’s a test of the full bandwidth capability of the connection. But this is not necessarily the actual results for cellphone users for a few reasons:

  • Cellphone providers and many other ISPs often provide a burst of speeds for the first minute or two of a broadband connection. Since the vast majority of web events are short-term events this provides users with greater speeds than would be achieved if they measured the speed over a longer time interval. Even with a speed test you often can notice the speed tailing off by the end of the test – this is the ‘burst’ slowing down.
  • Many web experts have suspected that the big ISPs provide priority routing for somebody taking a speed test. This would not be hard to do since there are only a few commonly used speed test sites. If priority routing is real, then speed test results are cooked to be higher than would be achieved when connecting to other web sites.

The Akamai numbers also can’t be used without some interpretation. They are measuring achieved speeds, which means the actual connection speeds for mobile web connections. If somebody is watching a video on their cellphone, then Akamai would be measuring the speed of that connection, which is not the same as measuring the full potential speed for that same cellphone.

The two companies are measuring something totally different and the results are not comparable. But the good news is that both companies have been tracking the same things for years and so they both can see the changes in broadband speeds. They also both measure speeds around the world and are able to compare US speeds with others. But even that makes for an interesting comparison. Ookla says that US mobile speed test results are 44th in a world ranking. That implies that the mobile networks in other countries make faster connections. Akamai didn’t rank the countries, but the US is pretty far down the list. A lot of countries in Europe and Asia have faster actual connection speeds than the US, and even a few countries in Africa like Kenya and Egypt are faster than here. My conclusion from all of this is that ‘actual’ speeds are somewhere between the two numbers. But I doubt we’ll ever know. The Akamai numbers, though, represent what all cell users in aggregate are actually using, and perhaps that’s the best number.

But back to my own cellphone, which is what prompted me to investigate this. Using the Ookla speed test I showed a 13 Mbps download and 5 Mbps upload speed. There was also a troublesome 147 ms of latency, which is probably what is accounting for my slow web experience. But I also learned how subjective these speeds are. I walked around the neighborhood and got different results as I changed distances from cell towers. This was a reminder that cellular data speeds are locally specific and that the distance you are from a cell site is perhaps the most important factor in determining your speed. And that means that it’s impossible to have a meaningful talk about mobile data speeds since they vary widely within the serving area of every cell site in the world.

The Fastest ISPs

PC Magazine has been rating ISPs in terms of speed for a number of years. They develop their rankings based upon speed tests taken at their own speed test site. They had about 124,000 speed tests taken that led to this year’s rankings. The scoring for each ISP is a composite number based 80% on the download speed and 20% of upload speeds. To be included in the rankings an ISP needed to have 100 customers or more take the speed test.

You always have to take these kinds of rankings with a grain of salt for several reasons. For example speeds don’t only measure the ISP but also the customer. The time of day can affect the speed test, but probably the type of connection affects it the greatest. We know these days that a lot of people are using out-of-date or poorly located WiFi routers that affect the speeds at their computer.

Measured speeds vary between the different speed tests. In writing this blog I took four different speed tests just to see how they compare. I took the one at the PC Magazine site and it showed my speeds at 27.5 Mbps down / 5.8 Mbps up. I then used Ookla which showed 47.9 Mbps down / 5.8 Mbps up. The Speakeasy speed test showed 17.6 Mbps down and 5.8 Mbps up. Finally, I took the test from Charter Spectrum, my ISP, which showed 31.8 Mbps down / 5.9 Mbps up. That’s a pretty startling set of different speeds measured just minutes apart – and which demonstrates why speed test results are not a great measure of actual speeds. I look at these results and I have no idea what speed I actually am receiving. However, with that said, one would hope that any given speed test would probably be somewhat consistent in measuring the difference between ISPs.

The results of the speed test ‘contest’ are done for different categories of ISPs. For years the winner of the annual speed test for the large incumbents has been Verizon FiOS. However, in this year’s test they fell to third in their group. Leading that category now is Hotwire Communications which largely provides broadband to multi-tenant buildings, with a score of 91.3. Second was Suddenlink at 49.1 with Verizon, Comcast and Cox and closely behind. The lowest in the top 10 was Wow! at a score of 26.7.

Another interesting category is the competitive overbuilders and ISPs. This group is led by Google Fiber with a score of 324.5. EPB Communications, the municipal network in Chattanooga, is second at 136.1. Also in the top 10 are companies like Grande Communications, Sonic.net, RCN, and Comporium.

PC Magazine also ranks ISPs by region and it’s interesting to see how the speeds for a company like Comcast varies in different parts of the country.

Results are also ranked by state. I find some of the numbers on this list startling. For instance, Texas tops the list with a score of 100.3. Next is South Dakota at 80.3 and Vermont at 70.6. If anything this goes to show that the rankings are not any kind of actual random sample – it’s impossible to think that this represents the true composite speeds of all of the people living in those states. The results of this contest also differs from results shown by others like Ookla that looks at millions of actual connection speeds at Internet POPs. Consider Texas. Certainly there are fast broadband speeds in Austin due to Google Fiber where all of the competitors have picked up their game. There are rural parts of the state with fiber networks built by telcos and cooperatives. But a lot of the state looks much like anywhere else and there are a lot of people on DSL or using something less than the top speeds from the cable companies.

But there is one thing this type of study shows very well. It shows that over the years that the cable companies are getting significantly faster. Verizon FiOS used to be far faster than the cable companies and now lies in the middle of a pack with many of them.

This test is clearly not a statistically valid sample. And as I showed above with my results from various speed tests the results are not likely even very accurate. But ISPs care about these kinds of tests because it can give them bragging rights if they are near the top of one of the charts. And, regardless of the flaws, one would think the same shortcomings of this particular test are similar across the board, which means it does provide a decent comparison between ISPs. That is further validated by the fact the results of this exercise are pretty consistent from year to year.

Speed Tests

cheetah-993774Netflix just came out with a new speed test at fast.com which is intended to measure the download speed of Internet connections to determine if they are good enough to stream Netflix. The test only measures the speeds between a user and the Netflix servers. This is different than most other speed tests on the web that also look at upload speeds and latency.

This raises the question of how good speed tests are in general. How accurate are they and what do they really tell a user? There are a number of different speed tests to be found on the web. Over the years I have used the ones at speedtest.net (Ookla), dslreports.com, speed.io, the BandWidthPlace and TestMySpeed.

Probably the first thing to understand about speed tests is that they are only testing the speed of a ping between the user and the test site routers and are not necessarily indicative of the speeds for other web activities like downloading files, making a VoIP phone call or streaming Netflix. Each of those activities involves a different type of streaming and the speed test might not accurately report what a user most wants to know.

Every speed test uses a different algorithm to measure speed. For example, the algorithm for speedtest.net operated by Ookla discards the fastest 10% and the slowest 30% of the results obtained. In doing so they might be masking exactly what drove someone to take the speed test, such as not being able to hold a connection to a VoIP call. Ookla also multithreads, meaning that they open multiple paths between a user and the test site and then average the results together. This could easily mask congestion problems a user might be having with the local network.

Another big problem with any speed test is that it measures the connection between a customer device and the speed test site. This means that the customer parts of the network like the home WiFi network are included in the results. A lot of ISPs I know now claim that poor in-home WiFi accounts for the majority of the speed issue problems reported by customers. So a slow speed test doesn’t always mean that the ISP has a slow connection.

The speed of an Internet connection for any prolonged task changes from second to second. Some of the speed tests like Netflix Ookla show these fluctuations during the test. There are numerous issues for changing speeds largely having to do with network congestion at various points in the network. If one of your neighbors makes a big download demand during your speed test you are likely to see a dip in bandwidth. And this same network contention can happen at any one of numerous different parts of the network.

The bottom line is that speed tests are not much more than an indicator of how your network is performing. If you test your speed regularly then a slow speed test result can be an indicator that something is wrong. But if you only check it once in a while, then any one speed test only tells you about the minute that you took the test and not a whole lot more. It’s not yet time to call your ISP after a single speed test.

There have been rumors around the industry that the big ISPs fudge on the common speed tests. It would be relatively easy for them to do this by giving priority routing to anybody using one of the speed test web sites. I have no idea if they do this, but it would help to explain those times when a speed test tells me I have a fast connection and low latency and yet can’t seem to get things to work.

I think the whole purpose of the Netflix speed test is to put pressure on ISPs that can’t deliver a Netflix-capable connection. I don’t know how much good that will do because such connections are likely going to be on old DSL and other technologies where the ISP already knows the speeds are slow.

Broadband Map of the US

broadband-by-congressional-districtConsider the following map of US broadband. This map was compiled by Gizmodo using data on broadband usage gathered by Ookla. This is a rather different map than the official US Broadband Map that is generated by the FCC. The official map uses data that is self-reported by the carriers. However, this map has been created by sampling and pinging actual Internet connections. Ookla owns speedtest.net and tests millions of connections from all across the country and at all times of the day.

This map is at a fairly high level and is shown per congressional district. But more detailed maps are available at the state and County level.

broadband-by-congressional-district

This map shows that there is a wide disparity of broadband speeds around the country. One surprising finding to me is that the average Internet connection is now at 18.2 Mbps download. The map then goes on to show those areas that are faster than average in blues and slower than average in reds. The 18.2 Mbps number is faster than I expected and goes to show that carriers around the country have been increasing speeds. This is certainly faster than the speeds that have been reported by other sources.

When you look deeper than this map at the broadband statistics you see a lot of what you would expect to see. Urban areas generally have faster broadband than rural areas. And the Verizon FiOS areas have much faster broadband than other parts of the country.

And this map shows some areas with fast broadband that might surprise people. For example, North and South Dakota have faster than average broadband. This is because the states are largely served by independent telephone companies that have built fiber into small towns and rural areas. And central Washington has some of the fastest broadband in the country thanks to several municipal networks that have built fiber-to-the-home.

One thing the map doesn’t show, at this high level, is that there are pockets of fast Internet scattered in many places. There are FTTH networks built in many small towns but these towns are not large enough to skew the data for the larger congressional districts shown on this map.

One state with high broadband is Florida, where I live. I have speeds available up to 104 Mbps from Comcast. The map for Florida shows what the cable companies are capable of and it’s a shame they have not improved their networks in more places to be this fast.

Ookla reports that the fastest town in the US is Ephrata, Washington with an average download speed of 85.5 Mbps. Second is Kansas City at 49.9 mbps. One would assume that with gigabit service that Kansas City will become the fastest place as more people are added to Google’s fiber.

One thing the map shows, is that an awfully lot of the country is below the average. Ookla reports that the slowest places are Chinla and Fort Defiance in Arizona which both have an average speed of less than 1.5 Mbps. These towns are within the Apache reservation and many native American towns are woefully underserved. The map shows large swaths of poorly served areas like West Virginia and Kentucky in Appalachia, like north Texas and Oklahoma, like Wyoming and Montana, and Maine.

I know at my house I have a 50 Mbps cable modem service and to me it feels just right. It allows us to watch multiple streaming videos while also working and using  computers for on-line gaming. I just moved from a place where my speeds would bounce between 10 Mbps and 20 Mbps and I can see a big difference. In my line of work I talk to people all of the time in rural areas who are still stuck with only dial-up or satellite as their broadband options. I know I could not do my job from such areas. These areas probably are not even showing up on this map because people who have connections that slow are probably not doing speed tests very often. They know they are slow.

The good news to me from this map is that the average speed in the US is up to 18 Mbps. But the bad news is that there are so many large areas left without good broadband. We still have a lot of work to do.