Hacker News Reader

Comment by jeletonskelly | Hacker News Reader

Imagine Amazon building and launching satellites like space based data centers. AWS in orbit. You could provision VMs on a satellite equipped with an array of general sensors, cameras, etc.

View on HN

Considering the difficulty of dumping heat in a vacuum and the cost of orbiting huge heat exchangers, I wager that heavy computing shall remain a mostly planetary activity for quite a while longer.

Isn't rather chilly space a great cooler for anything that heats up?

Sadly no. Space is cold but also very low on convenient mass to transfer the heat into. Vacuum is an ideal insulator. On Earth, you can put the heat into water, air, thermal pastes, etc., but in space, you have only slow old black body radiation, unless you do something more creative.

In simple terms space it not chilly but vacuum, so the heat can't go anywhere.

It's not that heat can't go anywhere, it's that there's nothing to carry heat away.

On Earth, air molecules can carry heat away from you. You transfer heat to the air around you. In space, there is almost no matter around you to absorb the heat and carry it away from you.

Your only option is to radiate heat away from you in the form of infrared light but that is a slow process.

> It's not that heat can't go anywhere, it's that there's nothing to carry heat away

So ... the heat stays put and doesn’t go anywhere?

It doesn't go anywhere except by radiation, which sucks as a way to cool things compared to conduction.

Not as well as intuition suggests. Deep space is cold, but also the sun (and reflected energy from the Earth!) is pretty hot. Without air, there's no convection so it's all radiated energy transfer, so you're limited by how much surface area you can "point at" deep space while not also pointing it at the sun.

The ISS has as much heat venting hardware as solar panels

Space near Earth is hot (apparently, ~ 120°C) in the sun, and cold in the shade (e.g., the Earth's shadow). On average, it's not particularly cold. You can get colder average temperatures on Earth, without the wide swings.

But more importantly, space is empty, and the best way to cool is to dump heat in some medium which carries it away (air is the basic one on Earth, but applications that really need cooling like to use water as a primary medium.) In space you've got...nothing, basically, somyou are stuck with radiating, which works, but poorly.

Space in itself has no temperature. It does have some background radiation, and thus you can measure its temperature (pretty cold at something slightly over absolute zero).

Here's the thing though. There's very little matter even in low earth orbit, so you can't use matter to transfer heat, as you would do on Earth (conduction and convection are out). That leaves only radiation, which requires a pretty large surface area. Try placing a computer on vacuum even on Earth and report back how the temperature looks like.

So no. Getting rid of heat is a big issue over there. Even in the shade.

Vacuums (or near vacuums) are actually good insulators. Double-walled insulated travel mugs are a great example of this.

Also, if you think about this in the context of the Sun, you get a really visceral feel for just how much energy is being produced. Black-body alone is transferring that energy to us. Scary.

Use a laser beam for heat dissipation. Bonus: Cloud Death Beam.

The SciFi novel "Sundiver" by David Brin is based around using a laser to dissipate heat from a manned spacecraft. Sadly thermodynamics mean it wouldn't work in reality.

http://www.davidbrin.com/sundiver.html

Heat is a problem, but I think it is safe to say that power is a bigger one. Those big arrays are really expensive, really heavy, degrade in output over time, and have been known to fail on occasion. [0][1]

[0] https://spacenews.com/33046spate-of-solar-array-failures-on-... [1] https://sat-nd.com/failures/index.html?https://sat-nd.com/fa...

Radiation is another issue. I have heard that radiation hardening is very costly

CDN on Starlink sorta makes sense

My first thought was DNS cache.

Genuinely curious - what would be a good use for a generic cubesat with average-ish components, like cameras, sensors, radios, etc. - nothing custom made and overly specialized? Especially considering the cost.

I would guess you'd have a relay for any data you want transferred from non-connected devices/locations, e.g. farm equipment, wind turbines, ships, etc. Also, a client-to-space-to-EC2-only VPN in a I-am-my-own-satellite ISP may have benefits. Unsure whether by "average-ish" you meant the ability to broadcast/receive ground data.

Is it really cost effective to roll your own for that kind of thing? I would have thought there were enough off the shelf satellite/m2m solutions for remote connectivity, eg https://www.orbcomm.com/en/networks/satellite

> Is it really cost effective to roll your own for that kind of thing?

Nope, hence leveraging Amazon et al. I would consider the AWS "above the cloud" offering (as I've now dubbed it) to essentially be the same commoditized thing as what you linked. An extraterrestrial CDN with edge compute if you will.

Clearly this is a case of "cloud in the sky"

This is my research area! (JPL)

Is Mark Watney a cool dude?

I only work with robotic astronauts.

so with the robot guy, you guys really don't need to send too many people to ISS, right?　

How different is a commercial setup compared to running a pi with rtlsdr and a homebrew helix antenna?

A ground station is going to involve some RF equipment first: an LNA to boost the received signal then that feeds into a down convertor to bring it to 1200MHz typically. The 1200MHz signal goes into a modem which can either be something very specific (think satellite TV set top box) or something very capable (think big server with a custom ADC capture card in the back). The modem will do some forward error correction. FEC can involve a mix of Reed-Soloman, Viterbi, and LDPC. The bits can now be sent out over a network connection. Uplink involves basically the reverse process except there is usually no FEC since you can always boost your ground antenna to be much more powerful than the one in space. An exception to that would be uplink signals meant to be relayed to the ground. More of those "bent pipe" satellites are becoming digital which allows them to process signals and make them smarter about channel utilization.

All of this equipment can usually fit within a rack or two. The downlink stuff can fit within a few U. Uplink involves some big power amplifiers. Back in the 60s it would have involved dozens of racks of equipment.

Similar to Azure talking about developing underwater datacenters

Did you know that eric schmidt funds a deep sea drone submersible project out of alameda to map and sensor the oceans... its funded under his and his wifes foundation, but basically its a secret project to map the oceans. They want to be first to own all that data

Got a citation handy by any chance?

It's called Schmidt Ocean Institute... not a huge secret.

The odds that the military already owns a lot of it (from operating submarine sonar and all of the things that come with it) are pretty high.

That project has been defunct for years. The company that claimed they could do it (and never even got started) has pivoted to Navy research projects.

A 501(c)(3) dedicated to open sharing of data is an odd vehicle for trying to competitively own data.

Eh. Space is just a place that's annoying to ship to. (With the one key advantage of having line-of-sight to arbitrary deployed IoT devices.)

"Underwater", meanwhile, is more of a mode of operation / design for data centers, with the goal of that simplifying (and decreasing costs for) cooling.

An underwater data center doesn't imply that the data center is actually off-shore to any large degree. I'd imagine they'd look more like undersea-cable landing points, built "into" a beach.

This will accelerate global warming and kill off fragile ecosystems from all the heat being radiated to the oceans.

I tried looking for environmental impact studies for Microsoft's data center project, but all I could find were object tracking models that their team built for tracking marine life.

If a published study on this exists, I'd love to take a look.

Maybe, but it could also reduce fossil fuels used to operate less efficient cooling systems in traditional data centers. I’d love to see a breakdown on the impact of each approach.

Not likely, water is a good heat conductor so there will just be a area of slightly warmer water near the outlet. You might get a micro ecology with some tropical fish far from their normal grounds but nothing dramatic will happen.

It'll certainly not be worse for the environment than any normal data centre, unless it springs a leak and something toxic escapes.

Underwater data centers make a great deal of sense - close to customers (low latency), fast deployments (decision to power on in 90 days), and, of course, cooling efficiencies. I expect them to actually happen if Phase 2 goes well.

Then again, I'm a bit biased - I worked on both Phase 1 and 2.

Im looking forward to the launch-rocket api.

What problem does this solve?

It would finally put the cloud in the sky! Marketing will be ecstatic over all the sky based puns they'll be allowed.

Also, I guess it would solve the problem of it being expensive to put things in space by virtue of somehow having made someone else pay for vast overcapacity of sensors.