[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Recommendation for Fedora server?



Les wrote:
On Mon, 2007-10-01 at 12:54 -0600, Karl Larsen wrote:
Kam Leo wrote:
On 10/1/07, Karl Larsen <k5di zianet com> wrote:
Bob Goodwin wrote:
Les wrote:
These supplies are switching supplies, not linear.  You should check
some of the draws, I think you will be surprised about how little some
of them draw (admittedly there are various design strengths out there.)

In general, a linear supply is about 20-35% efficient, meaning that a
450 watt supply would draw about 90 watts all the time.  But these new
switchers are about double that or more, meaning that the supply will
draw somewhere between 20 and 50 watts.  Moreover that draw will
basically be independent of the size of the ultimate supply, simply
existing to control the switcher itself, and not the drive current.
Technology exists in some forms of "buck control" that could boost the
on line efficiency further reducing the drain.  Power stuff is evolving
even faster than much of the other technology, but is less glamorous, so
doesn't get as much of the press.

Regards,
Les H


Here's some more interesting information on power supplies.  Note that
best efficiency is attained at/near maximum load.  So a 500 watt
supply looks good but may not be what you want when considering power
consumption and disposing of waste heat.

http://services.google.com/blog_resources/PSU_white_paper.pdf

Bob Goodwin

    Hi Bob, I am interested in what Google has done but need a lot more
information. They say changing the PS to one making just plus 12 volts
will save energy. I do not see that as a given. In fact it makes me
think they are just passing on the voltage change to the mother-board
makers. I think the idea being the MB makers can make just the other
voltages their board needs. But converting 12 to 1.5 volts with economy
of power is complex and will jack up the MB cost somewhat, and will make
it hotter.

Snip sig.
They're trying to get rid of distributing AC to all the racks and the
losses resulting converting from AC to DC for each MB.  If the MB
design only requires 12 VDC input then it is more efficient to have
only one AC to DC converter and buss the DC to all the boxes in a
rack. By the way, people have been bussing 24 and 48 VDC in equipment
racks for years.

The 12 volts can not be moved as 12 volts DC very far. The length of a car is about it. So Google must have a lot of racks in close proximity.

And while moving this single voltage around has lots of advantages for
racks, it is not such a clear advantage for single systems. I guess only time will tell us that
story.

What Karl has seen in working on Amatuer Radio, and my experience on a
boat tells us the story on high current busing which is a very complex,
very hazardous proposition.  And yes I know that it is done all the time
in racks, and along buses, however, check out the AWG specifications for
12v with 10% drop at 30' and think about the amount of copper needed for
a data center.  Up to about 5 racks is probably acceptable, but beyond
that the copper cost and the copper losses begin to mount quite rapidly,
especially for hard drives or other motor driven equipment.  Anything
that generates force, motors, relays, driver circuits for lasers, lamps
etc. force the current requirements quite rapidly.  In addition, as the
voltage drops, the area of the conductor goes up by the square.  This
establishes a point of optimum economic return, and also establishes the
acceptable connector loss and of course the resultant heat and fire
potential, which are primarily current dependent (a 500Watt soldering
gun actually uses only about 1.2V at its tip loop for example.)

	Then there is another whole subject that I touched on, that of power
factor.  Generally AC loads develop maximum power at 90degrees between
current and voltage.  But some components, such as totally passive ones,
develop maximum power and efficiency at 0 degrees difference between
current and voltage.  The variance between 0 and 90 degrees determines
the efficiency of power transfer.  So a very efficient device at 0
degrees may only be 30% efficient at 90 degrees and vice-versa.  However
if the lines are loaded with a 0 degree phase, the copper losses mount
and the wires carrying AC begin to heat up.  This increases losses
accelerating heating... a condition known as thermal runaway, and can
result in catastrophic failure of a power delivery conduit.  Not good.
So as the world matures, and develops more and more electrical loads, we
are all having to learn more and more about power and AC functionality
in the engineering community.  Lessons the RF guys have known for years
are now beginning to affect the digital and power folks and the colleges
are struggling to bring all this into the curriculum in a manner that
reflects good design practices as well.
	I spent much of my life working on RF, so I find it interesting, but
not too demanding to apply these principals to AC power.  Those without
a background in RF power electronics occasionally find themselves
struggling to get a handle on the implications.

Regards,
Les H

Hi Les the typical measure of by how much the current leads or lags the voltage is called "power factor". This is VERY important to power providers and they spend big bucks for Capacitors to put across the line to try and keep power factor = 1. If the power factor gets to around 0.5 and the power company is wasting a lot of power providing current that is wasted.

50 years ago I worked for Hughes and was put in charge of the MA1 fire control power supply for the F-106 airplane. I was told the 1600 Hz power source was running hot. I went to the factory to the system running and found the right cable and measured the power factor and found it was .06 about, inductive. I got a worker to put in a .8 UFD capacitor which when I measured pf again it was .95 and the system cooled down.

So for AC power performance you want the current and voltage to be in phase.



--

	Karl F. Larsen, AKA K5DI
	Linux User
	#450462   http://counter.li.org.


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]