[Beowulf] materials for air shroud?

Lux, Jim (337C) james.p.lux at jpl.nasa.gov
Fri Sep 16 09:39:40 PDT 2011



On 9/16/11 8:35 AM, "Prentice Bisbal" <prentice at ias.edu> wrote:

>Predicting airflow, especially in the turbulent regime, is not trivial,
>and not always intuitive. Just ask John Hearns about that. He works for
>McLaren's F1 team, and I bet they spend A LOT of computational power
>studying the aerodynamics of their cars (I'm sure the exact amount is a
>closely guarded secret).
>
>Luxi, I'm sure, could pipe in, too.

Unless you're like John with a F1 budget, or an America's Cup team, or
someone designing the next tactical fighter, airflow design in equipment
is done by skilled engineers using empirical experience, and a whole lot
o' prototyping.  Modern IR thermal cameras help, too.

I would guess that the vast majority of PC case design is done by old
fashion cut and try, and what modern technology has done is make the cut
and try faster: nifty ways to do the cutting, better tools to do the
trying.

Interestingly, since it's so "craft-like", I would think that there is no
reason a small integrator with a connection to a decent sheet metal
fabricator could crank out customized cases for a particular need: and if
the job were big enough to cover the cost of the iterations, you could get
a decent product.  All it takes is happening to have the right guy or gal
who has a good intuitive feel for it and some experience.  Think of all
those guys and gals doing intake manifold design, valve porting on
cylinder heads, etc. in the racing or performance car biz.

You find out about them by word of mouth and some people get good
reputations for having the "touch". Sort of like when I got my wife a
custom 7mm wet suit, I got a couple recommendations for "oh, you want to
have Jeannie do it", and subsequently (the suit turned out great) I find
out that *everyone* knows that Jeannie is who you go to, because she
started aeons ago, working with some other guy, who was the "goto" guy
back in the 70s.  There are lots of places to get a custom wetsuit in
southern California, but relatively few of them have the "touch" to make a
suit that is comfortable and usable (and there's some application
specificity.. I heard there's a guy up in Santa Barbara who is the Urchin
Diving Wet Suit god.. He's probably retired since the urchin business is
essentially gone)

So the question is "how do you find the gifted case designer who
intuitively understands airflow".. Maybe that will be David? (when he
gives up the comfortable life at CalTech, and sets up shop next to the
surfboard shaper and fashion designers, and hangs out his "Mathog's custom
high performance cooling PC cases" shingle)

The same thing applies in microwave and RF design.  There's all kinds of
tools out there to do the analysis, but a lot of good design is just
experience and knowing how to put the parts together.  So there are these
little companies that specialize in some fairly narrow specialty (e.g.
Krytar for diode detectors, Marki for mixers, Spacek for amplifiers,
Wenzel for oscillators).  The troubling thing is that when the one or two
people at the company who really know how to do things decide to retire or
die, the company will usually be bought up, but will be restricted to
making copies of things they've done before.  And it takes some years
before the word gets out.. "Oh, Bob's no longer there, so don't bother
going to them for a custom")

However, as Prentice points out in his anecdote, there's an awful lot of
"rack n stack" vendors out there who are just plain ignorant. And
ignorance is fine, as LONG as the buyer knows they're thermal design
ignorant.



>
>>From my experience, that's one of the factors that sets the large server
>companies, like Dell, HP, Sun, etc. apart from the smaller companies.
>These larger companies have the larger resources and the  engineering
>resources to model the airflow through their servers and it's effect on
>heat transfer.

I don't think they actually model, for the most part.  Sure, they'll have
some nice animations and some simple models they've run to put on the
sales brochures, but I'll bet most of it is empiricism.  And, in any case,
what you really need is the designer.. Whether they actually build the
prototype and test or build it virtually in the modeling code, it's not
like there's an automatic "thermal management system designer" program.

For instance, when we design radios for spaceflight, we start with some
notional design in terms of chassis wall thickness, number and thickness
of copper layers in the PWB, etc.   You run the model and see, Oh, that's
a 40C rise from baseplate to component, and since we have a component temp
limit of 85C and the baseplate needs to go to 55C, that's not going to
work.  Let's try making the copper layers thicker.. Run the model again.
Not quite.. Well, what about adding some support bosses to carry the heat
out from the middle of the board?  What about rearranging the parts on the
board to move the hotter parts closer to the edge? Oh no, the PWB designer
won't be able to fit the traces in., etc.

So it doesn't lend itself to a sort of "parametric" optimization.

You might do local parametric optimization:  Given this design, what if we
push 100cfm of air? What about 200cfm? Etc.



>
>Here's an anectdote about this issue:
>
>A certain national lab had incentives to use woman-owned and
>minority-owned businesses. So when they needed to buy ~200 cluster
>nodes, they went to a local small business that fit these criteria. I
>never saw the actual servers myself but sounded like the company bought
>generic 1U cases, and then put generic/commodity server components in
>them. I'm sure they didn't to any airflow or thermal analysis on the
>finished product.

I'd assume the same..
However, that doesn't mean that the small shop couldn't have done it, or
at least TESTED it.  Thermocouples are cheap.

That's what I find objectionable... It's really SO easy to do a simple
validation and test, but it's like
A) the procurement person doesn't know to put some sort of test
requirement into the requisition/RFP
B) the vendor doesn't volunteer to do it, or know how.

How hard is it to go through the mfr data sheet and find the recommended
maximum temperatures for the processor, then impose a requirement on the
vendor to demonstrate they meet it.
(well, actually, it *is* hard for the casual purchaser.. That's why the
successful cluster integrators earn their money.. They've either done the
work or they've at least done the tests and used that in their selection
process)


>



More information about the Beowulf mailing list