Broadband After the Pandemic

The New York Times is running an insightful article on things we’ve learned from the pandemic as they relate to future ones, 14 Lessons for the Next Pandemic. Given the importance the Internet has had to our response in terms of working, schooling, socializing, and shopping from home, some implications jumped out about making the Internet do even better next time around. And there will be a next pandemic.

In general, the Internet did well from a performance point of view. Several analyses confirm this fact, as the upcoming BITAG report will show. I’m not at liberty to share the report prior to publication, but look for it.

In general, Internet traffic increased by 30% or more and speeds held steady apart from Wi-Fi related congestion. While this was great for people who already had good broadband at home as well as good home computers, those on the wrong side of the digital inclusion gap suffered.

Work to be Done

We have work to do on getting people connected in areas that have adequate broadband service as well as in areas that don’t. These are two very different problems although we tend to lump them together under the label “digital divide” as if they were one and the same.

We also have work to do on the definitions of adequate as well as aspirational broadband services. The current official standard – 25 Mbps downstream and 3 Mbps upstream with limited latency – is still close to correct for most of the things we do on the Internet today.

The upstream standard is just barely enough to support two simultaneous video conferencing sessions. Zoom needs a sold one Mbps on the upstream side for its 1280 x 720 HD service, but it doesn’t leave much headroom for variations in computer and network load. Raising the upstream to somewhere between 5 – 8 Mbps would do the job.

Summarizing Needs

These two sources provide guidelines on the real requirements for today’s applications. unWired Broadband is a rural Internet provider in California’s Central Valley that uses wireless technology. They may have an incentive to underestimate application requirements, but their claims are consistent with my own measurements.

  • Netflix – 3 Mbps download for SD and 5 Mbps download for HD
  • Spotify – 96 Kbps download
  • Zoom – 2 Mbps download and upload

The FCC has also published its own estimates of application requirements for both individual users and households. Key findings:

  • General Browsing and Email – 1 Mbps
  • File Downloading – 10 Mbps
  • Streaming High Definition (HD) Video – 5 – 8 Mbps
  • HD Video Teleconferencing – 6 Mbps

The FCC’s general browsing figure is clearly out of date although it was updated just one year ago. Browsing is more like file downloads than email, so I would use 10 – 15 Mbps for browsing and 1 Mbps for email. Most of the email requirement comes from embedded images that may or may not be interesting.

For a more detailed look at browsing performance see my research paper You Get What You Measure; its old, but browsing is still browsing. 15 Mbps puts a single user in the highest performance class because websites are slower than networks.

Aspirational Standards

The rather modest error in the 25/3 standard has created an opportunity for several politicians (and regulators with political ambitions) to get attention for alternative standards. Four Senators ( Bennet (D-Colo.), King (I-Maine), Portman (R-Ohio), and Manchin (D-W.Va.)) wrote a letter to the FCC, USDA, Commerce, and NEC advocating a change to 100 Mbps down and also 100 Mbps up.

Going forward, we should make every effort to spend limited federal dollars on broadband networks capable of providing sufficient download and upload speeds and quality, including low latency, high reliability, and low network jitter, for modern and emerging uses, like two-way videoconferencing, telehealth, remote learning, health IoT, and smart grid applications. Our goal for new deployment should be symmetrical speeds of 100 megabits per second (Mbps), allowing for limited variation when dictated by geography, topography, or unreasonable cost.

The reasoning is weak, to put it mildly. The Senators don’t connect the dots between the applications they want so support with the bandwidth requirements of the applications.

The letter is also internally inconsistent. It argues that rural users need the same bandwidth as urban ones, which is fair enough. But it also admits that current broadband speeds in the US average “180 Mbps download / 65 Mbps upload with 24 milli-sec latency” per Speedtest.net.

Internally Inconsistent

The Senators want to use current broadband plan specifications as a goal only to reject them in favor of an arbitrary 100/100 requirement. This amounts to demanding nearly twice as much bandwidth for new rural builds as urban users have. Whatever that is, it’s not equality.

They also claim a connection between broadband service specifications with application requirements that simply doesn’t exist. As we have seen, 100 Mbps is 100 times more than consumers need for Zoom, the canonical video conferencing application.

Even if we increase the current Zoom-in-HD requirement to 4K, we don’t get past 5 Mbps. 4K pictures have a lot of information, but conferencing cameras don’t move and backgrounds don’t change, so we can easily compress their streams to no more than four times the current 1 Mbps requirement.

Poorly Researched

The other applications on the list don’t change anything: telehealth and remote learning have the same requirements as Zoom, and health IoT and smart grid are low bandwidth applications. The most demanding app in common use today, Virtual Reality, doesn’t even make their list.

Rather than doing the math on the sum of the application requirements, the Senators send us on a wild goose chase:

Ask any senior who connects with their physician via telemedicine, any farmer hoping to unlock the benefits of precision agriculture, any student who receives livestreamed instruction, or any family where both parents telework and multiple children are remote learning, and they will tell you that many networks fail to come close to “high-speed” in the year 2021. For any of these functions, upload speeds far greater than 3 Mbps are particularly critical. These challenges will not end with the pandemic. In the years ahead, emerging technologies such as cloud computing, artificial intelligence, health IoT, smart grid, 5G, virtual and augmented reality, and tactile telemedicine, will all require broadband networks capable of delivering much faster speeds, lower latency, and higher reliability than those now codified by various federal agencies.

The Senatorial logic seems to be something like “the more applications we can list, the more bandwidth is needed…because!” And the classification of 5G as an application that needs “broadband networks capable of delivering much faster speeds, lower latency, and higher reliability than those now codified” is just sad. It suggests that the staffers who drafted the letter simply went on a scavenger hunt for buzzwords.

Now for the Bad Part

As sloppy and inconsistent as this demand and its justification are, the really nefarious part is the letter’s exercise in forecasting. Not only must Congress dictate broadband speeds today, the letter suggests Senators want to dictate future speeds with an escalator:

Over the next five years, if current trends hold, data needs are expected to increase annually by at least 25% per year, according to the International Data Corporation. The goal of universal service requires that all Americans have affordable broadband with the technical capacity to meet those needs equitably.

Establishing a target today for all future requirements would be pretentious and arrogant even if the current goal were realistic. When the goal setting exercise is as sloppy as this one, it’s outrageous. Isn’t it enough to satisfy today’s needs today and leave tomorrow’s to the people of tomorrow?

One thing is certain: there has never been a time when users needed their data networks to be symmetrical. Campus networks (both wired and wireless) and many fiber optic services offer symmetrical transmit and receive speeds; but usage always follows an asymmetrical pattern regardless of capacity. In the real world, we use about 12 – 15 times as much downstream as upstream.

High Performance Hard Drives are not Symmetrical

No study has ever found symmetrical loading on the consumer side of the network connection. We see the same phenomenon in hard drives, where reads outnumber writes. The sensible way to design a data center SSD to avoid bottlenecks is to allocate more capacity on the read side than on the write side, as the new Intel drives do:

The Intel® Optane™ SSD DC P4800X/P4801X maintains consistent read response times regardless of the write throughput applied to the drive. Average read response times remain below <30μs while maintaining a 70/30 mixed read/write bandwidth of 2GB/s.

This means that the high speed interface between the SSD and the computer is asymmetrical, with 70% of total bandwidth allocated to reads and the remaining 30% to writes. Capacity is allocated this way because reads are more common than writes.

Bandwidth Specifications are Technical Standards

The specifications of the SSD interface are dictated by the nature of the workload, not by some aesthetic preference for counterproductive symmetry. Given than the main function of the Internet is to retrieve information from remote SSDs for users, isn’t it reasonable for the network to allocate its bandwidth the same way the SSD does?

This is the sort of question an engineer asks when tasked with devising a national definition of broadband. The politicians ask a different question: “what do the neighbors have?” It’s possible that engineers and politicians will come to the same numbers in the end, but unlikely.

Hence, we have expert agencies such as the FCC and NTIA to set standards in a technically rational manner, if for no other reason than to protect politicians from themselves. The standard for broadband in the USA should an asymmetrical pair of numbers reasonably related to application needs with the precise numbers calculated by the FCC rather than Congress.

First Lesson of the Pandemic

The United States and other nations met the challenge of the COVID-19 pandemic in a less than ideal way because we made too many decisions on a political basis that should have been made on the basis of evidence; scientifically, as it were. Political bias made for both overly broad lockdowns and premature relaxation of restrictions in many places.

Sometimes the evidence was too weak for correct decisions, such as the early advice that people didn’t need to wear masks because airborne transmission didn’t seem likely. We never recovered from that advice, even though it did allow healthcare to treat non-COVID-19 conditions as it normally does when the supply of PPE was constrained.

Recommendation One from the Times – Prepare for What We Can’t Imagine – underlines the need to collect data without being wedded to it, to be nimble, and to consider multiple scenarios about what the future will bring.

In broadband policy, this means rejecting the search for “future proof technologies” such as symmetrical fiber optic cable in favor of looking at what people actually need, which more often than not is asymmetrical wireless together with some asymmetrical wired. Today’s successful ISPs invest in both wired and wireless technology because that’s what it takes.

Be Smart and Organized

The Times also says Put Science First, Figure Out Who Gets Priority Treatment, Don’t Leave It Up to the States, and Stop the Mixed Messaging.

I think we’ve explained what science can teach us about broadband standards, the other two are easy to decode.

Networks need the ability to prioritize service to critical applications such as telemedicine, so net neutrality is a non-starter. Does anyone even remember why anyone thought it was a good ideal to treat applications as if they were all equally urgent and valuable?

And what happens when we leave contagion policy up to the states? In the pandemic it meant death rates per 100,000 ranging from 32 and 34 in Hawaii and Vermont to 269 and 252 in New Jersey and New York. States don’t have the resources to address technical subjects correctly, so they effectively throw darts blindfolded.

Nowhere is this more clear than with state laws on privacy and net neutrality.

Take Effective Measures that Don’t Screw People Over

Invest in the Numbers, Be Nimble in Providing Treatment, Don’t Let Race and Class Determine Who Lives and Dies, and Don’t Be Ageist aren’t just good ideas, they either are or should be the law. Like the pandemic, broadband is something that costs a lot of money, reaches everyone, and makes many, many other things possible.

We need to bring broadband to rural areas, hence we need good data about where it’s needed and a sound plan to pay for it. Problems like this that have been unsolved for decades cry out to new technology.

We may just have that with LEO satellite constellations such as Elon Musk’s Starlink and similar systems from his four competitors. If you want to build a greenfield network you may need to start by building a rocket.

Another option is terrestrial wireless systems using 4G and 5G technology, some variation on Wi-Fi, or CBRS. One of the big hurdles is spectrum licenses, which CBRS makes very inexpensive. The problem with greenfield networks is less technical than financial, and all of the wireless options are less expensive than wires.

Wireless systems are also less likely to discriminate against communities with low income or low affinity for broadband. Providers don’t need as many subscribers to cover startup costs with wireless than with wires.

Full Participation in Solving the Problem

Communities Need to Prepare Too, Stop With the Fringe Treatments, and Let Teenagers Be Teenagers are among the most relevant findings from the pandemic to broadband. Communities can help, especially in areas with no broadband, by supporting municipal networks built by electric coops. They have many of the skills needed for billing and maintenance, as well as access to poles.

Fringe treatments have plagued the pandemic response as some people continue to believe that home remedies or off-label drugs such as Hydroxychloroquine and Ivermectin are panaceas. In the broadband realm the fringe is populated by advocates for muni networks where multiple competitors already exist. There’s a strong case for spending public money on a first network, but no case at all for the fifth or sixth one.

Finally, teenagers have a lot to teach the rest of us about broadband. While grizzled old fiber bigots were still insisting that desktop computers and wired networks were the only way to go, teenagers stuck with their smartphones and apps and proved them all wrong. Don’t assume that the problems of tomorrow will be solved by the technologies of yesterday.

Prepare for Potholes and Roadblocks

Sadly, the pandemic has shown us that the international order isn’t as orderly as we would like by forcing us to confront The China Problem. Our neighbors have also let us down by refusing to mask up and behave responsibly, causing us to Look in the Mirror and See Who We Are as a people.

So we had to work around China by mapping the SARS-CoV-2 genome along with our Western allies, develop tests through WHO (and much more slowly through CDC), and develop our own vaccines through US/European cooperation.

Three of the four extant COVID-19  vaccines were developed in Europe, manufactured in the US and abroad, and financed by US companies and the US government: Pfizer, Johnson & Johnson, and AstraZeneca.

My number one frustration with the neighbors has been mask refusal, an antisocial behavior supported by sloppy research that failed to show benefits when all types of masks are lumped together without regard for diligent behavior. The broadband parallel is research that doesn’t ask crisp questions or analyze data carefully, such as the OTI Cost of Connectivity studies that consistently prove one theory about regulation and pricing as well as the opposite theory.

The New Normal for Broadband

The pandemic has forced us to face some facts and allowed us to learn others. We’ve always known that many people don’t subscribe to available broadband services, but we didn’t know that five million homes could have connected any old time until 2020, when they did.

We’ve also know that some people rely on wired networks while others like wireless better, but we didn’t know that 12 million homes were willing to drop wired for wireless until 2020, when they did. It’s pretty hard to reconcile these two shifts, of course.

And we’ve known cord-cutting was a thing for several years, but we had not seen a real surge until now:

The share of Americans who say they watch television via cable or satellite has plunged from 76% in 2015 to 56% this year, according to a new Pew Research Center survey of U.S. adults. Some 71% of those who do not use cable or satellite services say it’s because they can access the content they want online, while 69% say the cost of cable and satellite services is too high and 45% say they do not often watch TV.

While demand for broadband is growing, it’s also shifting to mobile services where the primary device is a smartphone. Dissatisfaction with Wi-Fi is part of this trend, but new spectrum allocated by the FCC for Wi-Fi 6e may have an impact on this sentiment. Generally, about ten percent of American homes upgraded their service in 2020, on top of free provider upgrades.

As the pandemic starts to fade, we won’t return to the old normal but we’ll reach a new normal with more broadband of all kinds, especially mobile, and less TV watching.  Against that background, the efforts of Congress to shore up the old normal are going to fail.

The new normal is wireless first. The sooner lawmakers realize this the better we’ll all be.