Recently, I was persuaded to take part in a Simon Wardley (of Wardley mapping fame) research group on the future of telecoms. I’m not a telecoms expert, but that wasn’t why I was asked. As well as an engineer I’m also a sci-fi writer, and Wardley reckoned I’d be interested in how our burgeoning telecoms expansion off-planet might change communications.
I’m a big fan of DoINg My OwN ReSEarcH, so I decided to go out and do some. Is space about to fundamentally change the internet?
In any discussion of space and society, there is a giant satellite network in the room.
In SpaceX’s own words, the Starlink system uses “advanced satellites in a low orbit to enable … high data rate activities that historically have not been possible with satellite internet”. The implication is that Starlink is indeed going to radically alter the internet, but is that true? What is the internet anyway?
Wired or tired?
The fundamental answer to “WTF is the internet?” is: wires, computers, and storage.
The “wires” (network) part is traditionally broken down into roughly three areas: the backbone, the metro, and the last mile. Sometimes the wires are literally made from copper wire, but more often they’re fibre optic cables or radio waves.
At each of these three stages, the level of aggregation decreases (i.e. they are carrying traffic for a smaller number of folk). The backbone is how the internet packets of millions of users travel long distances (for example, the backbone includes the undersea cables). Up until now, in modern times anyway, the backbone has been entirely fibre optic.
The metro bit is what criss-crosses urban centres carrying the traffic of many thousands of users, and is also generally fibre optic.
The last mile is the connection to individual users and might in reality cover fifty miles or fifty feet, it depends where you live. It could be physical or radio waves.
The truth is out there
No one knows how many computers (CPUs) currently make up the internet. Experts reckon a plausible estimate might be one hundred billion. For the scifi inclined, that’s about the same as the number of neurons in the human brain, or indeed how many stars there are in the galaxy.
For now, the majority of these processing units (plus the storage they talk to) are located in data centres, but DCs aren’t the only place they exist. There are an increasing number of “edge” mini DCs closer to the data sources or consumers. Plus, users provide their own edge compute via devices like smartphones or even IOT toasters. Processing at the edge gets you faster response times, and can also lead to less data needing to be passed around the internet, which is more efficient (see my previous article).
Is Starlink going to disrupt this picture of the internet?
What do we know about Starlink? The answer is, surprisingly little.
We understand it’s a bunch (i.e. thousands) of low earth orbit (LEO) satellites. At the moment, the satellites can’t talk to one another directly, but, Starlink claims, they will be able to in future using a custom protocol over laser links. The sats connect to the existing internet via specialised ground station transceivers. At the client side, they talk to user terminals (antennas cum wifi routers) that people install in their homes or offices. That all sounds like Starlink aims to play a role in the last mile and possibly metro stages of the network.
SpaceX has already signed a deal with Google to install ground stations on-site at Google data centres world-wide. They want to access the internet through Google’s own DCs, and it sounds like Starlink’s terrestrial hops will also happen through Google’s network, using the company as its backbone provider.
This presumably means that as a user, you could access Google services through Starlink without your data ever setting foot outside a Google or Starlink connection. Starlink provides secure, end-to-end encryption, and the old world ISP backbone won’t get a look in. Again, this appears to be metro and last mile stuff.
However, SpaceX has also signed a deal with Azure to connect Azure’s new modular edge DCs to their main ones, which is something you would normally associate with a backbone, and implies they reckon they can get backbone-level bandwidth over those satellites. That’s interesting…
So, could Starlink potentially replace all the “wires” of the internet: backbone, metro and last mile? Is that their plan?
There’s a significant obstacle in the way of an Elonet. Historically, space has terrible bandwidth.
Radio waves have always been relatively poor at bandwidth provision. All previous satellite networks have run up against the seemingly inescapable limitations imposed by physics. The viability of an Elonet comes down to how much aggregate bandwidth the Starlink array will be physically capable of offering (i.e. across all users). In the past, the answer for satellite systems has always been: nowhere near enough.
Stay in your lane
Wireless data transmission is phenomenally useful for some use cases (smartphones for a start) but up until now, it hasn’t been able to hold a candle to fibre, or even copper, in terms of handling very large numbers of individual connections. The issue is separating transmissions and stopping them from interfering with one another. In free space, too many people can hear you scream. If you do it over VOIP.
When it comes to interference, wires are efficient. Two can run next to one another without the messages in one interfering with the other. Free space transmission is much less good at that. Using a frequency to one user stops a satellite using it to talk to others nearby (how “nearby” depends not only on how good the antennas involved are, but how good other people's existing old antennas are, which is the cause of 5G interfering with radio altimeters on older planes).
Radio networks are limited by the number of discrete frequency ranges they have available to communicate with endpoints. Traditionally, if a satellite has to talk to lots of end points that are close together, it quickly runs out of frequency ranges that don’t interfere. That limits the number of users each satellite can service.
There are other bandwidth division mechanisms (e.g. time division multiplexing, or TDM) or you can have more satellites or a larger range of frequencies, but fundamentally this is a physical limitation. It’s possible Starlink will only ever be able to offer services to a fraction of the users the current physical internet covers. Maybe they never intend to offer more. Or maybe they have a secret plan for how to do it. Right now, we have no idea.
Grounded in reality
Even if Starlink never becomes the Elonet, there may be other benefits from having it as part of the bog standard internet. It’s a potential last mile for folk in underserved remote locations, which are too expensive to lay wires to. If there are already wires to a particular location, another path would still provide resilience. It has already been used to provide emergency connectivity to volcano and war victims. The home stations could also get fairly sophisticated and provide better edge compute and caching.
There are also potential security benefits. Whilst systems are hard to make completely secure, t it would at least be possible to create systems without Huawei routers, and, if you managed to smuggle a home terminal past the more physical firewall, You could also bypass the deep packet inspection of the Great Firewall of China. In the world of the future, you may be deciding who you trust more - big tech or state governments.
There is no doubt that in its current form, as described by Verge, Starlink is beset by disappointing connectivity due to issues like rain fade, and by device design niggles. Having said that, it’s really only the start, so I can forgive those. Once tens of thousands of satellites are up there, the story will probably be different. However, that leads to at least two new problems: astronomy and space debris.
For most of us, modern life in the form of light pollution has already put paid to a lot of stargazing. However, astronomers fear that Starlink together with the other satellite systems like OneLink and Amazon’s Project Kuiper will be the final nail in the coffin. There will be nowhere to get a clear view of the sky either optically or using radio telescopes. Similarly, there is a risk we’ll make the vital LEO zone a no-go area due to collision risks. Starlink is already the cause of 50% of near collisions, with only a fraction of its projected satellites in place.
But think of the amazing latency!
You might be thinking, but the latency! You’re right that the speed of light is the fastest we’re ever going to reach, so electromagnetic wave based comms are always going to be the quickest way to communicate (let’s not start bloviating about quantum entanglement). However, I’m going to remind you of your ‘O’ level classical physics: time = distance/speed. Increasing speed cuts packet transit time (therefore latency), but increasing distance offsets all of that.
If you want really low latencies you need systems that are light-based but also small. Being small also reduces the interference issues, because we’re talking about fewer users in that relatively titchy area. So, you can get both good bandwidth and low latency as well as the potential for decent aggregate bandwidth by setting up loads of discrete local systems. Combine that with edge processing so data doesn’t always have to transit the backbone and we can get ourselves some very snappy performance.
That is the promise behind 5G. The trouble with satellite systems is they are big. Not space big, but big compared to your office, warehouse, or Hong Kong city street, which is what 5G is targeting.
Of course, if you don’t care about supporting large numbers of users, you want your data to travel a long distance, and you have bags of $$$ (you’re a high frequency trader!) then there are already private microwave networks that can span large areas, but these are definitely not backbone. They are more like a last mile, or a limited metro, which happens to cross the Atlantic, but carries messages from a small number of endpoints. They can never scale to everyone. The aggregate bandwidth of those networks is limited and they are expensive to build.
The reality is that the speed of light limitation means latency is always going to be an issue for all communications, whether over fibre or electromagnetic waves. Fortunately, in most circumstances that doesn’t matter. Bandwidth is more important for most use cases except niche ones like high frequency trading, which bypasses the internet, and some gaming or augmented reality (AR). Gaming and AR use clever tricks to get around the problem by leveraging human psychology and the “smallness” ploy I was just talking about (e.g. 5G and/or the edge).
The final frontier?
Starlink is interesting, but in reality, the issues it faces and the ones it works around are nothing to do with space. The bandwidth scaling problems, as well as ones of interruptions by birds or trees or weather, are similar to those experienced by any system relying on satellites or even wireless. Fundamentally, low earth orbit is not exactly deep space.
Networking in genuine space has very different issues. As a wise man once said, it’s big. Really big. The speed of light is a comparative snail’s pace.
You may think it's a long way down the road to the chemist
The latency of Starlink’s Earth to satellite links will be 25-35 ms, which is similar to the latency the existing backbone runs at. When you’re communicating with Mars (~4-20 minutes latency) or even the Moon (1 second) then you have different problems.
For truly distant communications we need something like Delay Tolerant Networking (DTN). This new field of networking uses techniques such as store and forward, caching, and forward error correction (which aims to avoid asking for slow retransmits by, for example, sending multiple copies of the data). Starlink doesn’t need DTN because it doesn’t operate at space distances and therefore space latencies. Its problems are basically terrestrial.
Deus Ex Machina
I’m not saying Elon Musk has a God complex, but I’m guessing he believes he’ll come up with some ultra-cunning wheeze that will overcome the interference issues of physics and get Starlink’s bandwidth high enough to take over the world.
Simon Wardley is betting on Starlink’s eventual satellite count being way higher than we are imagining, and he might be right. Will that be part of the bandwidth solution? In addition, Starlink is already attempting to grab a huge range of frequencies to operate over. France recently delayed the French rollout while they assessed the impact of the range Starlink was asking for.
Musk has promised to resolve Starlink’s astronomy and debris risks but, again, they seem tricky issues to fix. “Man’s reach should exceed his grasp,” is probably written on his business card.
Is it all worth it? Fundamentally, I reckon the answer to “Is space about to fundamentally change the internet?” is that right now, Starlink and its ilk are just more wires - if with some whizzy new features. Good, but not revolutionary and with plenty of risks and downsides. Perhaps, however, I merely lack faith. Or as Douglas Adams might say, I’m not asking the right question.