-Tom Lidikay


 

 

There has been some talk from small start-ups about putting a datacenter in space, for ultra-secure data-storage, and even some talks of virtual servers running on XenServer


Humans have been building all manner of technology to put into space for the past 60 years, so it should come as no surprise that we have gotten pretty good at it. But when putting stuff in space, there are special considerations, and when building a datacenter, there are other, also special considerations.

Being a humble computer engineer, I will not pretend to be an expert about spaceflight, but I will put my experience with computers to work with this list of things:

Consideration Number One: Radiation hardening


All things that go into space are being constantly bombarded by radiation that isn’t being blocked by the earth’s atmosphere, and magnetic field.
Of major consideration, are Neutrons, Protons, Alpha Particles, Heavy Ions, and very high energy Gamma Photons.

Gamma photons in Low earth orbit are mostly diverted and absorbed by the magnetic field and atmosphere, respectively, and Alpha particles are not great at punching through materials, so most outer casings will block them. But neutron, proton, and heavy ions are still a consideration in processor and storage chip design.

The consequences here are two fold, one being Lattice Displacement, causing actual physical damage to transistor junctions (Processors are made of millions of them) and Ionization Effects which can develop charges that create glitches, set and unset bits in memory, and can cause unexpected behavior.

Chipsets have been developed that are resistant to this radiation, from companies such as BAE Systems


The primary issue this presents for a low earth orbit datacenter, is that processor chipsets in a radiation hardened format are still primarily using a 45 nanometer construction.

When discussing processor construction, the nanometer comes up a lot. This is referring to the smallest cut that can be made when etching transistor junctions into a piece of silicon wafer. So, the smaller the cuts you can make, the smaller each transistor will be, and the more of them you may fit on a piece of given silicon.

Intel has been manufacturing 22 Nanometer processors for some time now, and odds are the computer you are reading this from uses that sort of construction. They made publication for 14 Nanometer construction in 2015, and are now talking about 7 nanometer to come.

As far as datacenters go, running a virtual server on a 45 Nanometer processor, would be a huge setback. Not even mentioning what kind of investment it takes if you want to be picky about which architecture you want. Most radiation hardened chipsets are made around an ARM architecture, as of the time of this writing, I had trouble finding anyone making an x86 chipset.

Consideration number two: G-forces

Space may be a harsh place, but getting there is a little tricky too.

When going to space, the massive amount of acceleration generates additional forces upon everything you put up there. For sattelite launches, every component that goes into the craft is rated at about 6 G’s.

That is 6 times normal earth gravity. As discussed above, there are chipsets designed to withstand these forces, but they are incredibly different from what most datacenters are running.

If you are the intended customer base for the “flying virtual server” then this may not bother you as much, but if you have a special need, then this wouldn’t be right for you. If you want a graphics chipset, or an ASIC, or anything that doesn’t fly with the rest of the server hardware, then you are on your own.

Consideration number three: Maintenance and service

Given it is my line of work, along with many other professionals, it’s no small secret that computers break a lot.

Most of the time stuff can be done remotely, but other times you are going to need a hands-on fix. Digital systems can handle most of the innocuous “press the button on the front of the machine” stuff, but what about bad ram chips that need removing and replacing?

Any hardware failure that requires hands-on to fix means flying a mission specifically to the orbiting payload to fix it. Which is very expensive.

The other alternative is to shut down that part of the system and abandon it permanently, which is what a lot of satellite payloads end up doing. Most of the time, these are set on a course towards earths atmosphere, where they will burn up on re-entry. But its an incredible waste of money when talking about what could be an entire server rack or two worth of hardware. Throwing away an entire rack of functioning servers because of one bad component would be absurd.

Low earth orbit satellites move, very very fast. The average satellite in LEO is moving at 28,080 Kilometers per hour, or 17,448 MPH. This means it will go around the entire planet, once every 90 minutes. So you can imagine, setting up a stable communications path with it is a lot of work.

At the very least, you need a communications platform that can track with the satellite as it moves past, and preferably multiple communications platforms at different places around the globe. This could be advantageous to you, if you want to use a resource in a different place than the one in which you created it. But for most, this is annoying. Can you imagine, trying to keep a file upload stable, when your endpoint keeps shifting around the globe? Most data protocols get extremely fussy about changes in their network path.

Geostationary orbits are much higher up than low earth orbits, and come with their own issues, like the ones above. More fuel is required to reach this height, and more radiation is present.



Consideration Five: Orbital debris


Putting anything into orbit presents a big uncertainty. Due to the velocities involved, there are a huge amount of objects that are travelling in different directions, at different speeds, as well as natural debris coming from meteors outside of earths orbit.

One of these can strike your datacenter at any given time.

Here is the ESA writeup about hypervelocity impacts.



A better plan:


Here is an idea of mine, just for giggles: what if we put a datacenter on the moon?

Lets address the issues I have brought up which are fixed by this:

Radiation:

The surface of the moon is still bombarded by the radiation, but we can dig under the moons surface to hide from it.

To quote a research paper entitled “Lunar soil as shielding against space radiation”


“For this beam, the average percent dose reduction over all soil
samples is 0.8%, compared to 0.9% for aluminum. This similarity is
not unexpected: for Apollo sample A70051, for example, the
weighted average mass number of the constituents is 26.3 and the
weighted average atomic number is 12.85, compared to nominal
A
and
Z
for aluminum of 27 and 13, respectively. In other words, per
unit areal density, lunar soil is only slightly less effective than
aluminum, and only about half as effective as polyethylene at
reducing dose for this particular ion and energy. Of course, soil has
the great advantage of being available
in situ
on the Moon.

Moon soil is almost as effective at blocking radiation as aluminum! This is great news! Only a few feet of soil would be more than enough to block almost all of the radiation. Tunneling channels 6 feet below the lunar surface into (presumably) solid rock would make construction of data centers much more simple.
Smaller configurations of processor are possible, since they will be run for their extended lifetime in this radiation shielded environment. It may even be possible to run off the shelf components, if transportation needs and costs permit, leading into:

G-forces:

Since we are no longer completely reliant upon running the processor chipsets, servers, etcetera, in its flight configuration, we can pack it up differently to withstand launch stresses. This could be anything from foam packing, like you might find in an ordinary server box new from the manufacturer, or even more advanced methods.

Service:

If we are already establishing a datacenter of some size in this scenario, its even feasible to have someone on-site fulltime to perform maintenance tasks. It would no longer be a matter of possibly throwing away hardware, it could be repaired, salvaged, or recycled. If we needed additional personnel, parts, etc. it would be a longer distance, but feasibly, those personnel could perform multiple tasks in one trip.
As a stretch goal, it could even be possible to establish this datacenter as a beginning project of developing outposts on the moon, even developing habitation centers, research facilities, and other things. At that point, a dedicated trip is no longer required. You just have to hop on the next trip out, along with the other cargo and passengers.

This one is a bit of a mixed bag. Getting a signal to the moon is difficult, just because of the massive amount of space in between, the One way path loss at 1296 Mhz is 206 Db. (Radio signal path loss can be determined using: Path Loss = (4πR)2/λ2 Where R is the distance to the moon and λ is the wavelength) A laser communications system would be the most ideal method, offering a significant amount of bandwidth. This will require some engineering work, but is entirely feasible given current technologies.

There are improvements to be had, as the moon just does not move that much. With an orbit time of 27 days, it is unlikely to break any single uploads or downloads, and nearly half the globe can see the moon at any given time. (“see” in this instance meaning it is still line-of-sight to the earths surface, regardless of whether it is lit by the sun or not at that time,)

Fewer upload stations are required, since single stations can perform more of the upload/download work. The end point will no longer shift across the global internet infrastructure frequently, and basic fail-over routing will work just fine to handle these duties.

Orbital Debris:

Because the moon is tidally locked with the earth, there is far less risk of a meteor strike on the moons earth facing side than its outward facing side, or in orbit. Being up and out of the man-made objects orbiting the earth, there would be virtually no risk of a collision with orbital debris. And, any object that does come near it, given our underground construction, is very likely to be stopped by the soil, causing no damage. Destruction of the datacenter is much less likely, which is a good thing for our data.



So, in conclusion, we should put the datacenter on the moon, not in orbit. That isn’t to say there are no valuable things we can put in orbit, for example , the plan for internet distribution is an interesting idea.