noise Re: [Beowulf] Re: Setting up a new Beowulf cluster

[Jim Lux]

At 02:16 PM 2/13/2008, Mark Hahn wrote:
>>>Your cluster will need a home, and there are good homes and not
>>>so good homes depending on its scale.
>
>
>>too much noise.  Did I mention NOISE?  There is no inexpensive way to
>>quiet down a rack because to first order sound insulation == heat
>>insulation. \

Actually, no.. good acoustic isolation is not good thermal 
isolation.  Sure, things like fiberglass batts provide thermal 
insulation and also (slightly) attenuate high frequencies.

What you want (for acoustics) is loss, mass, 
compliance/springyness.  That is, the sound wave hits something, and 
rather than being transmitted somewhere else, it just dies 
there.  The traditional approach is to build something like a 6" 
thick wall with alternating 4" wide studs (i.e. no direct mechanical 
connection between the panel on one side and the panel on the 
opposite side of the wall).  In between the gap, one hangs thin 
sheets of lead, or plastic loaded with metal filings, bituminous 
rubber (aka Bituthane, a synthetic rubber sheet loaded with asphalt 
for mass) etc.

There are these nifty things called "Z brackets" that are used to 
attach the wallboard to the studs in acoustic isolation walls.  They 
help a lot, because they isolate the vibration of the panel from the stud.
(Google found this: 
http://www.wilrep.com/WilrepDataSheets/noisecontrol/RSIC/Clips.htm)

So, you have acoustic wave on one side hits wall.  Wall is hopefully 
not a stiff diaphragm, but is something moderately compliant, and 
lossy (imagine hanging a carpet up), so its movement dissipates some 
energy.  The back side of the wall transmits an acoustic wave to the 
sheet of lead.  It doesn't displace very far, and is quite soft, so 
it just deforms and gets warm. (one would not, for instance, want a 
thin beryllium or titanium membrane.. things that make good speaker 
cones make terrible acoustic isolators).

And so on.  It's basically a cascade of lossy low pass filters 
(inductance = mass, resistance = mechanical loss, capacitance = 
springiness... you want big L and C (low frequency cutoff) and you 
want big R (lots of loss)


Your big sound transmission paths are going to be unavoidable 
mechanical connections (the floor, or sill plate of the wall) and 
unbroken air paths (door gaskets, a/c vents, etc.)

For windows, you use two panes, set at an angle relative to each 
other, of different sizes, in resilient mountings... this reduces the 
coupling from one pane of glass to the other.





>putting a dozen 15k rpm fans in each node is the noisiest possible way
>to move the air.  a few very large fans could be quite quiet, and 
>not involve much in the way of either sound or heat insulation.

Indeed.. the way to move lots of air is big fans turning slowly (and 
think how post-apocalypic sci-fi movie it looks.. Think of movies 
like "Total Recall".  A guideline in the HVAC business is to keep the 
velocity of the air below 1000 feet/min, preferably half that.

And, there are HUGE differences in the noise level for fans with 
identical air moving performance. The amount of mechanical energy 
going into acoustic power is tiny, even if the sound is quite loud.

A helicopter radiates, for instance, about 10 milliwatts of acoustic 
power, and moves on the order of a million cubic feet per minute of air.


>there are some instances of using heatpipes to get heat out of the node
>and into chilled water or refrigerant.  it's a little unclear to me why
>this is not done more, especially among the blade-loving community.


Maintenance hassles.  some of the larger IBM S/360s were liquid 
cooled, and a pain to deal with.  Anything with liquids and 
connectors will leak.


>then again, a quiet cluster is less impressive when it comes time to 
>impress the visitors ;)


That's what the big Tesla Coil or quarter shrinker is for.

>_______________________________________________
>Beowulf mailing list, Beowulf at beowulf.org
>To change your subscription (digest mode or unsubscribe) visit 
>http://www.beowulf.org/mailman/listinfo/beowulf