Monday, July 13, 2015

Orbital Data Centers

Cloud computing is the idea of storing data on a server or having that server perform tasks so there is less load on your computer. But these servers are located in warehouses on Earth. They require large amounts of energy, they have a large footprint, and they have to transmit data over a distance to your computer.  What if server farms were placed into orbit? What if cloud computing occurred above the clouds?

Orbital server farms would have several benefits to their terrestrial counterparts. The first is power. In orbit there is an abundant supply of solar energy, twice as much as what enters our atmosphere and hits the ground. It is also possible to be in continuous sunlight. This provides a clean and inexhaustible power supply for the farm.

Next, there is an abundant amount of space in space. A server farm may grow as large or larger than any building on Earth with.

Last, concerning transmission time, space really is halfway to anywhere. Today and in the future satellite transmission will be common. The trip up and then back down takes only millionths of a second but that is very slow for a computer which works with billions of operations per second. Halving the distance of transmission by just sending information down, instead of up then down, would increase internet speeds.

But that is a far rosier picture than reality. While endless solar power is available in orbit it would still require huge arrays to power a server farm. The ISS solar array is the size of a football field and provides 110 kilowatts of energy, enough to power 55 houses. But a 55,000 square foot server farm on Earth uses 5 megawatts or enough to power 5000 homes. Clearly a development in space power systems will need to be built. But computing is becoming more efficient all the time so such power requirements may not remain standard.

Next, indeed a server farm may grow as large as it wishes in orbit. But with increased size comes increased weight and therefore higher launch costs. The cost of space launches would have to become a thousand times cheaper than the cheapest SpaceX launcher existing today in order to compete with terrestrial installment. However initial launch costs may be offset if lifetime power costs were lower due to orbital solar plants.

Last, while internet and communication speeds may be twice as fast by placing the servers in orbit this may not be a significant enough trade-off for the risks of creating such a facility. But Google and others are working on using satellites to provide internet connectivity to the world, adding a few servers to the satellites is not a great leap.

But perhaps the biggest problem of all to such facilities may be cooling. Servers become very hot and require constant removal of heat. But in space there are few options for heat removal. Heat can only leave by radiation. But a satellite using solar power is also constantly absorbing heat as thermal radiation. If a satellite also has a hot server farm at its center then a substantial thermal management problem arises.

Despite those challenges here is why such data centers will exist. Just as Google and Facebook are trying to provide internet connectivity to places where there is none, so to do they need to provide data storage and computation. Second, having large power hogging facilities in orbit will reduce the load on Earth power plants and not be in danger of blackouts while using the Sun's energy. Last, if humans go to Mars they will need computing power. But it makes no sense to attempt to land servers on Mars when they can perform just as well in orbit and perhaps provide services to multiple settlements. The same goes for the Moon. The effort of landing a sever farm on another body is to great when it can just be parked in orbit.

The latter scenarios may be where orbital computing systems begin. Providing computing resources to future Mars missions. Small server farms requiring nothing on the scale of terrestrial data centers could be created and then launched to support space missions. This is something that already has many engineering precedents in places like the ISS. It a project which is well within current technologies to achieve and could actually be built by a small firm or just a few people in a garage. From there such orbital data centers could follow a modular growth plan where new, self-contained modules are continuously combined to create ever large server farms which are able to finally provide meaningful quantities of computation to the world or other colonies.

Data centers in space are something which have potential and are currently feasible. However in order to scale them to compete with existing data centers in a substantial way requires growth and construction of other space resources. However, small scale orbital server farms can be created to support missions to other planets which will be coming in the next decades. Whether they will ever replace their cousins on the ground on Earth is unknowable until the industry develops more.