[Beowulf] Why We Need a Supercomputer on the Moon

Lux, Jim (337C) james.p.lux at jpl.nasa.gov
Tue Oct 2 11:50:54 PDT 2012


On 10/2/12 8:56 AM, "Eugen Leitl" <eugen at leitl.org> wrote:

>Why We Need a Supercomputer on the Moon
>By Robert McMillanEmail Author 10.02.12 6:30 AM
>A Lunar supercomputer could ease our deep-space networking bottleneck
>Simon Lutrin/Wired
>Should we build a supercomputer on the moon?
>It would be a mammoth technical undertaking, but a University of Southern
>California graduate student thinks there¹s a very good reason for doing
>It would help alleviate a coming deep-space network traffic jam that¹s had
>NASA scientists worried for several years now.

It's not as bad as all that.  ACtually, some folks in the DSN are worried
about a drop-off in business base.  In any case, you can have as much data
throughput as you are willing to pay for. The limit is one of physics, and
for going from "deep space" to Earth, or Moon, or Earth Orbit, or Lagrange
points, the distance is about the same.

You want to send X bits per second back to earth, you just figure out how
much energy you need to put into each bit, and now you know what your
radiated power needs to be; write (a big) check, and you're done.

(Telecom is NOT usually the big part of a mission budget, by the way...A
300 million dollar mission might spend 10M, total, on all the telecom
hardware, testing, etc.)

Doubling the data rate requires doubling the power, and that can get
expensive (solar panels aren't cheap, or, if you want a bigger antenna to
reduce the beamwidth, again, not cheap, and limited by the size of the
launch vehicle shroud).

IN any case, a few years back, I did a design for a >100Mbps link back
from Jupiter, and there was nothing particularly bold or unobtainium about
it, except the dollar cost and the electrical power requirement.

A bigger problem is that the spectrum at the deep space X-band allocation
(8400-8450 Mhz) is only 50 Mhz wide, and there's a lot users in one place
(Mars) so the spectrum is full.   However, there's plenty of bandwidth
(500 Mhz) up at Ka-band (32 Ghz)..

>Ouliang Chang floated his lunar supercomputer idea a few weeks ago at a
>conference in Pasadena, California. The plan is to bury a massive machine
>a deep dark crater, on the side of the moon that¹s facing away from Earth
>all of its electromagnetic chatter. Nuclear-powered, it would process data
>for space missions and slingshot Earth¹s Deep Space Network into a brand
>moon-centric era.
>³Once the physical infrastructure backbone is laid out, I suspect it would
>look much like the monolith excavation site in Clarke and Kubrick¹s 2001:
>Space Odyssey,² says Chang¹s course supervisor Madhu Thangavelu, of USC¹s
>Viterbi School of Engineering.

With choral works by Ligeti playing in the background?

>The Deep Space Network is a network of 13 giant antennas located in the
>Australia, and Spain that gather data and talk to spacecraft in, well,
>space. These space missions are already fighting for bandwidth on this
>overloaded network and most of the data has to get back to Earth for
>processing. With a lunar supercomputer, Chang says, that could change.

I don't know that "fighting for bandwidth" is right.  What they do have to
do is negotiate for "time on the antenna", because the current user model
is one user at a time. However, you could do multiuser comms (from say,
the moon, or Mars) by just adding receivers, not by adding antennas.  And
the new "advanced receivers" being built these days would do this by
changing software.

>His supercomputer would run in frigid regions near one of the moon¹s
>The cold temperatures would make cooling the supercomputer easier, and
>it to use super-efficient superconductive materials to move around
>electricity. Although it¹s not clear how much water could be found on the
>moon¹s poles, Chang envisions a water-cooled supercomputer.

I would have a hard time believing it's easier to reject heat by radiation
to cold space than by conduction to water in a convenient river or lake,
or to air, pretty much anywhere.  Moving heat from one place to another is
the expensive part and would need to be done whether lunar or terrestrial.
 Disposing of the heat is easy.

>How much would this Lunar supercomputer cost? Well, Chang and Thangavelu
>it costs about $50,000 per pound to ship materials to the moon. Add to
>the cost of digging out and building out the sub-lunar supercomputer
>cooling system and nuclear power generator, and you can easily envision a
>project in the $10 billion to $20 billion range, never mind the cost of
>building a lunar base station. That would easily make it the most
>supercomputer ever built.
>The lunar computer would communicate with spaceships and earth using a
>of inflatable, steerable antennas that would hang suspended over moon
>craters, giving the Deep Space Network a second focal point away from
>Chang¹s Lunar supercomputer complex Image: Ouliang Chang
>Some at NASA agree that there¹s a coming Deep Space Network traffic jam.
>in 2006. the agency¹s top networking gurus warned that over the next three
>decades there will be an ³order-of-magnitude increase in data to and from
>spacecraft and at least a doubling of the number of supported spacecraft.²

And, in fact, this is true.  MSL/Curiosity is sending back about a Terabit
per day, which is a lot more than MER sent back (MER was designed for 8-10
kbps via X-band direct to earth).  Over the first 90 sols, MER sent back
about 200-250 Mbits/sol via relay, and maybe 20-30 Mbits/sol on X-band.
(By several months into the MER mission, X-band DTE usage had dropped off
dramatically).  The relay orbiter (e.g. MRO) sends back the relay data at
Mbit/second rates (it has a big antenna and a big amplifier).

So, yes, Terabits is "order of magnitude" increase in less than 10 years.

>Space scientists are worried that the existing Deep Space Network
>hardware is
>obsolete and just not up to the job of transmitting the growing workload
>extra-terrestrial data.

Well, yes.. They are worried, but not so worried they're willing to spend
any money to fix it.   It costs about $50M to build a brand new 34m
station. The big 70m antennas were originally built as 64m antennas for
Apollo, and definitely are showing their age.  However, they cost several
billion back in the 60s.

There's all kinds of low cost ways to do downlink: arraying multiple
antennas for instance.  The Allen Telescope Array is sitting mothballed up
in Northern California, and could easily be pressed into such service,
fairly cheaply.  

The going rate for downlink from commercial vendors is on the order of
$1000/hour (for instance, you can rent the Stanford dish for about that

>The U.S. space agency is going to have to come up with a plan, the
>In fact, Chang isn¹t the first person to propose putting a big data
>processing facility on the moon. Back in 2004, researchers at Space
>Loral described something called the Lunar Data Cache ‹ an
>backup system that would keep businesses online in the event of a Sept.
>2001-type terrorist strike somewhere on Earth. The Loral proposal also
>described a few way-out moneymaking ideas such as lunar rover-deployed
>billboards, robotic rock-heaving contests, robot wresting, and rover races
>piloted by NASCAR drivers.
>Clearly, the business of dreaming up supercomputers in space is not for
>who think small.
>Still, after being reported in the New Scientist, Chang¹s work has caught
>interest of the space community because it addresses a very real and
>space problem. And that¹s something that could give future lunar missions
>very clear and exciting sense of purpose, says his USC¹s Thangavelu. ³For
>now, it is piquing the mind of the policy makers,² he says.
>³Even though far-out,² Chang¹s paper ³does excite the imagination,² says
>Bhasin, a system engineer and formulation manager with NASA Glenn Research
>Center. He agrees that there is a growing problem with outer space
>Bhasin works on designing advanced space networking technologies, and he
>that there are some promising alternative communications technologies ‹
>Laser-based networking, for example ‹ that could break the Deep Space
>Next year, researchers at MIT¹s Lincoln Laboratory will test out a 622
>megabits per second Laser network that will speed up earth to moon
>communications fivefold.
>The supercomputer-on-the-moon idea was pretty much a natural for Chang,
>describes himself as a super-computing geek. ³My PhD thesis is about doing
>one of the largest space plasma turbulence simulation in the world,² he
>³I just put two popular concepts ‹ space exploration and cloud computing ‹
>So, will there ever be a supercomputer on the moon? Though NASA¹s Bhasin
>finds it interesting, he¹s really not sure it will ever happen. ³Your
>is as good as mine,² he says.  Robert McMillan
>Robert McMillan is a writer with Wired Enterprise. Got a tip? Send him an
>email at: robert_mcmillan [at] wired.com.
>Read more by Robert McMillan
>Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
>To change your subscription (digest mode or unsubscribe) visit

More information about the Beowulf mailing list