[Beowulf] Why We Need a Supercomputer on the Moon

[Lux, Jim (337C)]

On 10/15/12 4:09 PM, "Tomasz Rola" <rtomek at ceti.com.pl> wrote:

>On Tue, 2 Oct 2012, Lux, Jim (337C) wrote:
>
>[...]
>> 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.
>
>Cool. It would be nice to have kind of live transmission from next probe
>somewhere there. All those Jupiter ascents and descents, watched from a
>"window inside a monitor". Almost like being there, minus pesky AI trying
>to freeze me out of the ship and stupid####### benevolent aliens trying
>to 
>lure me into journey to nowhere...

I will assure you that we do not have an AE35 unit in our antenna control
logic, failed or not.  Interestingly, though, an optical tracker isn't a
bad idea for this kind of thing.. Earth is very bright (unless it's
between you and the Sun, in which case tracking the Sun works..).





>
>> >Space Odyssey,? says Chang?s course supervisor Madhu Thangavelu, of
>>USC?s
>> >Viterbi School of Engineering.
>> 
>> 
>> With choral works by Ligeti playing in the background?
>
>Yeah, I can bet they plan to send a chorus there, too.
>
>Those guys talk like this is a five year long project, after which all is
>set and running. But in reality, especially nowadays-like reality
>(including budget constraints and lack of political will), I guess the
>shortest time to have it is more like 30 years. If all goes well.
>
>BTW, I think I never have been a big fan of sending supercomp up there.
>At 
>least not after I learned that space is nasty place to live without good
>shielding. I guess this haven't changed recently? Also, CPUs being sent
>there are actually special radiation-hardened versions AFAIK, and I think
>much more pressure is put on their reliability than speed.

Interestingly, modern standard parts tend to be pretty rad tolerant.  To
make them work with such tiny geometries, they have to be very heavily
doped, so even if you dump a bunch of charge carriers in with a cosmic
ray, it still doesn't change the behavior.  The other thing is that with
the geometry being so small, the probability of hitting any one device is
small.

Shielding helps with total dose and long term damage (bias shifts,
leakage, etc.), but not with high energy Galactic Cosmic Rays or high
energy particles from the Sun.

A lot of the designs now have built in redundancy or error detection and
correction for other reasons (e.g. RAM with EDAC).




>So they are not 
>quite good as supercomp building blocks and average PC is no good when
>being zapped by ultrafast particles every few minutes or exposed to CMEs
>from the Sun...


Not as bad as all that.  I've actually been looking for an actual
documented instance of a commercial non-space part failing in space.
There's quite a few flying, but nobody has been able to point to a
specific case.

The most troublesome parts these days, in a permanent damage sense, are
FETs in DC/DC converters: lots of voltage stress across them, so they get
Single Event Gate Rupture.
>
>BTW2, maintaining this behemoth would be a (logistical) nightmare if it
>ever got built. I mean, it is ok to send people to the Moon and live in a
>permament base for purposes other than taking care of poor man's copy of
>Google computing container, full of hardware designed in tech process
>generations behind the newest one (thick paths last longer)...
>
>BTW3, I love(d) the story of maintaining/debugging Deep Space remotely:
>
>http://www.flownet.com/gat/jpl-lisp.html
>
>Just my random thoughts on the subject. Will be happy to be corrected by
>someone.
>
>Regards,
>Tomasz Rola
>
>--
>** A C programmer asked whether computer had Buddha's nature.      **
>** As the answer, master did "rm -rif" on the programmer's home    **
>** directory. And then the C programmer became enlightened...      **
>**                                                                 **
>** Tomasz Rola          mailto:tomasz_rola at bigfoot.com             **