[Beowulf] Re: blackbox on Mars?
James.P.Lux at jpl.nasa.gov
Fri Oct 20 07:17:58 PDT 2006
At 06:52 AM 10/20/2006, Robert G. Brown wrote:
>On Thu, 19 Oct 2006, Jim Lux wrote:
>>At 03:56 PM 10/19/2006, steve_heaton at iinet.net.au wrote:
>>>G'day Jim and all
>>>Pushing further from the topic (given your day job Jim ;) Also
>>>this is pure speculation...
>>>30% efficiency on the solar cells?! A generous round up I suspect? I thought
>>>Deep Space 1 only got ~23%? And that was the soopa doopa focus
>>>layer, no dust
>>>out there to speak of flavour?
>>You betcha.. by the time you can put 4-5000 kg of container on the
>>surface of Mars, I suspect that 30% solar cells will be a
>>reality. Actually, Mars Science Lander (MSL) for launch in 2009 is
>>probably pretty close to 1000kg (I'd have to go check, but it's
>>widely described as being roughly the size and mass of a Mini).
>Has anyone remarked on the absurdity of sending a cluster to the surface
>of Mars under any cirmcumstances?
I think that's what the discussion is about. Sun posted a picture of
a "blackbox" sitting on Mars with a couple solar panels on the top
behind one of the MERs. I did some back of the envelope and figured
that it probably was out of the box by orders of magnitude.
> What in the world would it do that
>couldn't be done far cheaper back on earth, or in orbit?
Anything that requires realtime control with a time scale smaller
than the propagation delay of the comm path.
Things like real time image processing for navigation (although
that's not cluster scale, yet.. they do it on a much slower machine,
but, on the other hand, they don't drive all that fast
10cm/sec). Several DARPA grand challenge contestants had small
clusters in their vehicle for this kind of thing though.
Signal processing for seismic or subsurface radar is another possibility.
> There are damn
>few things that I can think of that might even NEED a cluster's worth of
>compute power on that end of things -- maybe, just maybe, running AI
>robots or helping to preprocess images -- but the former SEEMS like it
>could be managed with a very small cluster -- 4-8 CPUS tops, mostly for
>failsafe -- and the latter could be done anywhere.
>Moving data is cheap.
Not from Mars, and particularly not from the surface of Mars. You'd
have to trade the cost of processing on the surface (which would
allow a smaller, lighter, lower powered radio) against the cost of
processing elsewhere (cheaper) but that would require a bigger
heavier radio with more power.
> In fact, it is "free" -- available at opportunity
>cost on a communication channel that has to exist no matter what, and
>bandwidth in that channel is similarly cheap, all the way to Earth.
That channel is not cheap, nor is it available 24/7. For instance,
you might be on the far side of Mars, with no direct path to
Earth. There are a few relay satellites available (most with very
skinny pipes) and for the current ones at least, the comm pass lasts
<20minutes. Putting a 24/7 communications infrastructure in place
around the moon or Mars is a pretty expensive proposition (many
billions of dollars). It might well be that spending a billion
dollars on a 1000kg cluster on Mars is cheaper than 10 billion
dollars to create the broadband communications infrastructure needed
to put it elsewhere. Part of the problem is the way in which money
for such things is doled out.. generally by the missions, which are
science driven. Human spaceflight is a notable exception for the
money distribution, and infrastructure (if needed for the human
flight) can get funded. It also helps that human flight is, overall,
a bigger chunk of change than robotic flight (robotic exploration
missions run in the $100M to $750M range, and tend to stand alone,
although part of an overall program)
>only thing that is obviously limited in the channel is latency -- order
>of an hour to Earth, order of 0.001 sec to various possible orbits
>One MIGHT consider sending a cluster to sit in a geosync orbit AROUND
>Mars with a high bandwidth uplink (literally) from the surface and order
>ten millisecond latencies. That's fast enough to permit the cluster to
>drive a rover or do other AI stuff and still beat human reaction times
>of 100 msec. It also permits it to get power from LARGE solar panels or
>a LARGE reactor (relatively speaking, with the usual, fairly easily
>solvable, engineering problems associated with dumping heat in a
>vacuum). It is so much cheaper a solution than dropping something so
>extraneous and useless to the actual surface that nobody sane would EVER
>consider the latter.
You'd be surprised... for batch processing type applications, where
it's ok to wait til the next satellite pass for the next data bolus,
the trade is easy.
> That's if they don't decide to skip the not
>inconsiderable expense of shipping a cluster to Mars orbit and ship the
>DATA all the way back to Earth, where clusters are cheap and easy to
>maintain and don't have to be failproof to five nines for a decade in a
We prefer the term failure tolerant. (fail proof is hard to come by)
But there is a fair amount of work on how to do just that, and
clusters are actually a good way to have the *system* reliability be
good, even if the *component* reliability isn't. (barring aging and
systemic failure mechanisms.. total dose damage, for instance)
As above, there is a tradeoff
>Just a thought... so regardless of the "corporate hype" about sending
>ANY sort of cluster "to Mars" only a complete idiot would seriously
>contemplate sending one to the surface, and only a RICH idiot (something
>oxymoronic in that, no?) could afford it.
Lot's of rich naive folks around.. Someone has to pay for all those
expensive Arabian horses and F1 race cars and 200 ft yachts.
SO far, no patron has stepped up with $200M to send something to Mars, yet.
James Lux, P.E.
Spacecraft Radio Frequency Subsystems Group
Flight Communications Systems Section
Jet Propulsion Laboratory, Mail Stop 161-213
4800 Oak Grove Drive
Pasadena CA 91109
More information about the Beowulf