Q&A: Emerson’s Jack Pouchet explores with CBR the limitations and benefits of launching data centres into outer space.
As data centre consolidation continues on Earth, industry visionaries are exploring different concepts to help computing reach as many people as possible.
Jack Pouchet, VP for market development at Emerson Network Power, came up with the idea of data centres in space while working on rockets and other space technology in Palmdale, US in the early 2000s.
He said: "As Man moves further into space, we are going to take data centres with us. It seems like a natural evolution of the [data centre] industry. It is going to happen. Who is going to be the first one to experiment in space? We have already been doing it."
Pouchet said that for example, humans are already putting computers in space, "we are putting the functionality up there, we are using it for the intended purpose, but we have not said that the intended purpose now could be to use it as a data centre".
"Maybe somebody is already trying this and we do not know," he said.
CBR: How would you do computation in space?
JP: Keith Lofstrom, in 2009 wrote a paper, Server Sky, and his idea was putting computation in space, not the building. With space, you put [the system] in the right place, you got unlimited power.
Our idea was to put [the data centre] in orbit, deep enough where you can take advantage of daylight 24/7. That puts you a bit further out so it changes the model of what you are going to be computing and the way you operate it, but you can work on the continuous duty cycle and work through the issues of latency.
CBR: What are the limitations to the idea of data centres in space?
JP: Limitations are the fact that you are away from the Earth. You are not going to use computing in space for real time financial transactions, for example.
Things where latency is an issue, you do not want to work with, but maybe for big data analytics and some other computational models, even some business models that we have not even developed yet, having a resource that could be used around the clock and maybe having distributed connectivity to the Earth, could be a fascinating resource.
CBR: How much would such idea cost?
JP: By the time we work through the launch model, we would be able to work through the server model. It is not going to take a full framed server in the way it is built today. We are going to have to work out the details of what that server looks like, what that storage system looks like up in space.
In terms of the launch cost, the good news is that that is rapidly declining. What might have been $100 million a few years ago, by the time we are ready to tackle this [server] problem it could be down to $10 million. Elon Musk and his organisation, and a host of others that are privatising space, are rapidly bringing the launch cost down.
Looking at Moore’s Law, computational workloads are getting cheaper and faster, so how do we, as an industry, take both of these trends and take advantage of the unlimited energy potential by putting something in space.
CBR: How would you finance the projects?
JP: One area where this seems to have some potential are the undeserved and emerging markets. The obvious downside to that is that it costs an awful lot to get it up there.
If you are going to put something in space, you are going to have to spread the cost over the widest user base you can, or have someone that is going to spend a fortune in it.
Microsoft, Google, AWS, Alibaba, etc, seem to be likely candidates to participate in that space program. There are a lot of innovative ideas going on [like Facebook internet drones], which have different cost models and capabilities. I would suspect that others are certainly looking at this.
Astronaut carrying out maintenance on the ISS
CBR: How would maintenance be carried out?
JP: That is the biggest issue. There is the Server Sky approach: micro servers, effectively the size of a silicon wafer. Whatever the size of the wafer you are building with, the silicon wafer becomes the energy source, it becomes a PVU [processor value unit] wafer.
When you launch that into space you do not do maintenance. This model fits with people like Microsoft, Google, Amazon, etc. They run this very large distributed computational global frameworks. If at any one point in time there is around 1,000 servers failing for example, if you deploy one million servers, you have ten to 100 that would fail at any one day. Do you just run around and replace them all? No.
You wait until there is a critical mass that have failed and then you go out there and you do the service, or maybe you just let them all fail and replace the whole building. That would be a model in space, the depreciating asset.
CBR: What sort of security concerns are involved and how would you protect these assets?
JP: People are very concerned about that, including storage and [data] integrity, what we call the "Snowden effect".
If the data is in outer space, one would have to go up there. And when someone is on the way up there, I might just crash that asset back in the Earth, burn it and it is all gone.
The way it is going to be secured is whith frequency, some sort of cybersecurity code (that could be updated often). Unfortunately, you can only operate over so many frequencies.
Once you put a satellite in space, everybody knows the frequency that these things tend to communicate on. Maybe we will have another band communication system.
CBR: Would data centres be run with the laws of the country from where they were launched, or would an international space data centre regulation be created?
JP: That is a good point, and that is why we have the UN and we have several space treaties, such as no one can own the Moon. It would probably be like the International Space Station.
In the beginning we go to the old days, like when AT&T launched a satellite and owned it, subject to the laws of the US because there is where it was launched from.
We are going to see some of this changing. Different countries have different rules and some countries will try to say "you built that in Germany or in France, you launched it in Azores or in the US". Different people are going to try to claim rates on it. However, once it is up in space, it becomes a different asset.
CBR: Could data centres in space become a new business for collocation providers?
JP: Once they move beyond science experiments, it becomes the ultimate cloud. Is this going to be an Azure, Google, AWS?
[As for colos] there is Digital Realty Trust and imagine it becoming Space Realty Trust, who knows? Maybe they come up with a new model, or Rackspace really becomes Racklessspace. It lands itself to new business models and innovation.
NASA’s concept of a human colony on the Moon
CBR: Following outer space, where else could data centres go – the Moon?
JP: When we get to commercial travel it will have to happen [the deployment of data centres on the Moon]. You will need multiple data centres even to sustain life up there.
Local data centres are going to be required to run the air management systems, security, infrastructure space base power systems, and others. It is going to be one of the first things we will have to do when we start going to the Moon [and even Mars].
Data centres have to follow human life. They will probably not be massive buildings, but there is certainly going to be data centre requirements wherever we go.
CBR: Is Emerson doing any work around data centres in space?
JP: We are working with partners looking at new ideas. The companies mentioned, [Microsoft, Google, etc] we are in discussions with them and others. When someone is ready to make a move, we will be ready to support with any resources we can for the application.