[Cialug] [omaha] Google Datacenters

[Daniel A. Ramaley]

What i recall from physics and electronics classes is that how the power 
flows through a cable depends on the potential difference (voltage). At 
low potential, the whole cable is used. But at higher potential, only 
the "skin" is used. That partially explains why it is possible to 
survive being hit by lightning--most of current flows through a struck 
individual's skin and clothes, and not so much through their brain, 
heart, or other vital organs. My understanding of the reason that only 
the surface of the conducting medium carries much current at high 
potentials is that the repulsive forces of the electrons themselves is 
strong enough that they "want" to escape the medium they are travelling 
in. So, they move to the farthest points from each other, aka the 
surface of the wire (or object struck by lightning). It has been 
several years now since i've studied physics, and the flow of 
electricity was certainly not my area of expertise within that domain, 
so my understanding and explanations now are probably only marginally 
better than a layman's.

What was the potential (voltage) on that DC line in Montgomery Wards? 
How about in Hoover Dam? I'd bet in both cases they were rather high 
voltages. 12V for a computer is quite low and will require very thick 
cables to carry much power. P=IE. Power = current * potential. So, as 
potential drops, to carry the same power current has to increase 
proportionately. And higher current requires thicker wires. This is 
also why long-distance power transmissions are done in the 100kV to 1MV 
(or more) range--at 120V, 240V or even 10kV the wires would just be too 
thick. At low voltages (where the skin effect isn't significant) the 
cross-sectional area of the wire determines how much current it can 
carry. At high voltages the surface area is more important. 
Unfortunately i don't have a quick "rule of thumb" to tell at what 
voltage one needs to consider cross-sectional area vs. surface area, 
and it has been too long since i've studied the relevant equations to 
hazard a guess. I've read that some electrocutions are done at 5kV, so 
probably 5kV is still low enough for significant current to flow 
through the medium rather than on the surface.

Note that much telecom equipment runs at 48V DC, but 48V is 4x higher 
than 12V. So, wires with only 1/4 the cross-sectional area can be used 
to carry the same power. If you're talking about wiring a data center, 
that's a big difference in the amount of expensive copper that is 
required.

But unless there's an electrician or a physicist or some other 
knowledgeable individual in the house, i don't know how much more we 
can say about this. I'd like it if data centers could be run off of low 
voltage DC--it intuitively feels like a more elegant solution than 
having an AC/DC converter in each machine. (But then again, intuition 
in areas of physics is often incorrect...) Since PCs need more than 1 
voltage though, i'm not sure how much is really gained by changing the 
1 voltage they are fed from 120V AC to 12V DC; either way some extra 
conversion steps have to be done in the box. Even if it is fed 12V DC, 
and 12V is one of the voltages the machine needs, that 12V would have 
to be filtered in the box to make sure it is smooth enough for delicate 
electronics. I'm sure 12V is possible (i actually have a few low-power 
servers at home that i built which take 12V DC and use a "PicoPSU" to 
convert it), i'm just not sure such a low voltage is an efficient way 
to get power to a rack of high-power servers. Once someone actually 
does it, then the wisdom of the idea should be more apparent.

Maybe we should just switch PCs to require a 15kV AC power feed (aka, 
from a neon sign transformer). Data centers would have a fresh ozone 
scent like after a thunderstorm, and people working on the machines 
would learn very quickly to be careful and not make mistakes with the 
power.

On 2009-04-11 at 22:32:04, Daniel E Sloan wrote:
>As I recall from a power class...at this point many years ago...the
> electron flow is almost entirely on the outer skin of the cable. 
> This is why the main lines from the Hoover Dam are hollow.  (saves
> weight, while allowing maximum transfer)
>
>My grandfather was an electrician at the Montgomery Wards building in
> Chicago when all the power transmission from the utilities was in DC.
> So power transfer through a DC bus in a server room should certainly
> be feasible.
>
>Dan Sloan
>
>
>----- Original Message -----
>From: "Daniel A. Ramaley" <daniel.ramaley at drake.edu>
>To: "Central Iowa Linux Users Group" <cialug at cialug.org>
>Sent: Friday, April 10, 2009 11:08:23 AM GMT -06:00 US/Canada Central
>Subject: Re: [Cialug] [omaha] Google Datacenters
>
>Possible? Of course. Practical? I don't know.
>
>You need very thick wires (thick inch-thick bars of metal) to carry
>enough current at 12V if the distance spaned is very far. If the wires
>are too thin, they will have too high a resistance and drop too much
> of the potential. If the wires are really too thin, they will get hot
> and melt. Making thick enough wires out of copper might be
> prohibitively expensive. Cheaper metals that don't conduct as well
> would require even thicker wires. Of course, if you have a room
> temperature superconductor (presumably made of unobtainium), then it
> might be more feasible. But i don't know much about superconductors;
> they might have limits on how much current they can carry too.
>
>People with more electrical knowledge please correct me if necessary.
>
>On 2009-04-10 at 10:49:09, jrnosee at gmail.com wrote:
>>I've often wondered if it would be possible to convert AC to 12VDC
>> for the entire datacenter then run 12V lines to each server and only
>> have a DC->DC 12V/5V/3.3V Point of Load power supply do the
>> conversion. Thus eliminating power loss in the multiple AC-DC
>> conversions and running all the extra power supply fans, etc.
>
>----------------------------------------------------------------------
>-- Dan Ramaley                            Dial Center 118, Drake
> University Network Programmer/Analyst             2407 Carpenter Ave
>+1 515 271-4540                        Des Moines IA 50311 USA
>_______________________________________________
>Cialug mailing list
>Cialug at cialug.org
>http://cialug.org/mailman/listinfo/cialug
>_______________________________________________
>Cialug mailing list
>Cialug at cialug.org
>http://cialug.org/mailman/listinfo/cialug

-- 
------------------------------------------------------------------------
Dan Ramaley                            Dial Center 118, Drake University
Network Programmer/Analyst             2407 Carpenter Ave
+1 515 271-4540                        Des Moines IA 50311 USA