|
In the 2020 Broadband Deployment Report, the FCC made the following claim. “The vast majority of Americans—surpassing 85%—now have access to fixed terrestrial broadband service at 250/25 Mbps”. The FCC makes this claim based upon the data provided to it by the country’s ISPs on Form 477. We know the data reported by the ISPs is badly flawed in the over-reporting of download speeds. Still, we’ve paid little attention to the second number. The FCC cites—the 25 Mbps upload speeds that are supposedly available to everybody. I think the FCC claim that 85% of homes have access to 25 Mbps upload speeds is massively overstated.
The vast majority of the customers covered by the FCC statement are served by cable companies using hybrid fiber-coaxial technology. I don’t believe that cable companies are widely delivering upload speeds greater than 25 Mbps upload. I think the FCC has the story partly right. I think cable companies tell customers that the broadband products they buy have upload speeds of 25 Mbps, and the cable company’s largely report these marketing speeds on Form 477.
But do cable companies really deliver 25 Mbps upload speeds? One of the services my consulting firm provides is helping communities conduct speed tests. We’ve done speed tests in cities recently where only a tiny fraction of customers measured upload speeds greater than 25 Mbps on a cable HFC network.
It’s fairly easy to understand the upload speed capacity of a cable system. The first thing to understand is the upload capacity based upon the way the technology is deployed. Most cable systems deploy upload broadband using the frequencies on the cable system between 5 MHz and 42 MHz. This is a relatively small amount of bandwidth that sits at the noisiest part of cable TV frequency. I remember back to the days of analog broadcast TV and analog cable systems when somebody running a blender or a microwave would disrupt the signals on channels 2 through 5—the cable companies are now using these same frequencies for uploading broadband. The DOCSIS 3.0 specification assigned upload broadband to the worst part of the spectrum because before the pandemic, almost nobody cared about upload broadband speeds.
The second factor affecting upload speeds is the nature of the upload requests from customers. Before the pandemic, the upload link was mostly used to send out attachments to emails or backup data on a computer into the cloud. These are largely temporary uses of the upload link and are also considered non-critical—it didn’t matter to most folks if a file was uploaded in ten seconds or five minutes. However, during the pandemic, all of the new uses for uploading require a steady and dedicated upload data stream. People now are using the upload link to connect to school servers, connect to work servers, take college classes online, and sit on video call services like Zoom. These are critical applications—if the upload broadband is not steady and sufficient, the user loses the connection. The new upload applications can’t tolerate best effort—a connection to a school server either works or doesn’t.
The final big factor that affects the bandwidth on a cable network is demand. Before the pandemic, a user had a better chance than today of hitting 25 Mbps upload because they might have been one of a few people trying to upload at any given time. But today, a lot of homes are trying to make upload connections at the same time. This matters because a cable system shares bandwidth both in the home, but also in the neighborhood.
The upload link from a home can get overloaded if more than one person tries to connect to the upload link at the same time. Homes with a poor upload connection will find that a second or a third user cannot establish a connection. The same thing happens at the neighborhood level—if too many homes in a given neighborhood are trying to connect to upload links, then the bandwidth for the whole neighborhood becomes overloaded and starts to fail. Remember a decade ago that it was common for downloaded video streams to freeze or pixelate in the evening when a lot of homes were using broadband? The cable companies have largely solved the download problem, but now we’re seeing neighborhoods overloading on upload speeds. This results in people unable to establish a connection to a work server or being booted off a Zoom call.
The net result of the overloaded upload links is that the cable companies cannot deliver 25 Mbps to most homes during the times when people are busy on the upload links. The cable companies have ways to fix this—but most fixes mean expensive upgrades. I bet that the cable companies are hoping this problem will magically go away at the end of the pandemic. But I’m guessing that people are going to continue to use upload speeds at levels far higher than before the pandemic. Meanwhile, if the cable companies were being honest, they would not be reporting 25 Mbps upload speeds to the FCC. (Just typing that made me chuckle because it’s not going to happen.)
Sponsored byIPv4.Global
Sponsored byCSC
Sponsored byDNIB.com
Sponsored byVerisign
Sponsored byRadix
Sponsored byVerisign
Sponsored byWhoisXML API
A while back I wrote a blog item entitled “Why You Shouldn’t Believe Network Speed Tests”
The point made there is that there are many factors in play, ranging from packet maximum transmission unit to variation of the inter-arrival time between packets (i.e. “jitter”), to the burst pattern of packet drops. And each application, and each implementation of that application, has its own reaction to these kinds of things.
Most “speed testing” software is based only on the most simplistic of measurements. One would never consider the “0 to 60mph time” of an automobile as fully describing the capabilities or performance of that vehicle. Yet that is largely what we get out of “speed tests”: a single number that purports to represent everything.
In order to better evaluate the capabilities of a given internet link we need to measure many different attributes, ranging from packet/frame sizes to base latency and variation (jitter) to the burst characteristics of loss, reordering (yes it does happen, in switches/routers and also on some media that is formed out of a bundle of channels), and duplication (which also happens). Our experiences with bufferbloat also tell us in no uncertain terms that the size and nature of queuing within those links can have massive impacts on perceived performance.
I am keenly aware of the impact of a lot of these aspects because I build tools that create those effects, on demand, so that builders can evaluate whether their products work well under less than perfect conditions. (Over the years I have seen many products wobble and react poorly - or simply utterly collapse.)
(By-the-way at my home, Comcast’s top-tier service delivers a maximum upload rate of only about 11.5mbits/second over a Docsis 3.0 cable link that was tested and cleared of noise just a few days ago.) That’s not even half of the 25mbits/second number you mention.