[LUAU] Anyone use a Linksys NSLU2? USB hard disk drives are FOSSfriendly?

[Jim Thompson]

>> Brian Chee wrote:
>>> My biggest bitch about "normal" machines is that I want my NAS up  
>>> all the
>>> time and a 350watt power supply could potentially add something like
>>> $20/month to my electric bill....
>>
>> As a partial aside - if a power supply is rated at 350 watts, does  
>> this
>> mean that it constantly pulls 350 watts?  If so, I may need to do a
>> costs-benefits analysis of my own pc based file servers.

Simple switched power supplies employ a simple full wave rectifier  
connected to a large energy storing capacitor. Such a power supply  
draws current from the AC line in short pulses when the instantaneous  
line voltage exceeds the voltage across the capacitor. During the  
remaining portion of the AC cycle the capacitor provides energy to  
the power supply.

The problem is that charging or discharging a capacitor may cause  
energy loss even if no dissipative elements are apparent.

Consider two capacitors and a switch:

V=VI               V=0
       ------/   ------
       |                  |
C1=                 =  C2
       |                  |
       ----------------

Q=V*C

V=VI/2            V=VI/2
        ---------------
       |                  |
C1=                 =  C2
       |                  |
       ----------------

E=(C*V^2)/2

The top figure shows a capacitor C1 charged to voltage Vi and no  
voltage on capacitor C2 before switch closure.

C1 is equal to C2 and the energy in the system is:   Energy =  
(C1*VI^2)/2

After switch closure (bottom figure), the charge and voltage is  
divided equally between the two capacitors (conservation of charge)  
and the total energy in the system is: Energy = (C1*(V1/2)^2)/2 + (C2* 
(V2/2)^2)/2 = (C1*V1^2)/4

But wait, half the energy has disappeared. Where did it go?

The energy lost in directly switching voltage to a capacitor at  
another potential is lost in parasitic resistance, and if the  
resistance is too low, in arcing or welding of the switch contacts.   
Potentially some is also lost to radiation.

Resistance. The easiest loss mechanism to show analytically is the  
loss in parasitic resistance, such as the capacitor equivalent series  
resistance (ESR) or wiring resistance. Adding this resistance to the  
circuit and calculating the power dissipated shows the energy loss.  
The energy loss is independent of the value of the resistance.

Arcing. Most switches used in power supplies are solid state and  
arcing is not a problem, but if a capacitor is charged through a  
contact, arcing may be a problem.

Radiation. High rates of change of voltage or current result in  
radiation. Directly switching voltage to a capacitor at another  
potential is a potential source of radiation (EMI).

Note that the losses are the same if C1 is a voltage source instead  
of a capacitor.

Switching supplies also feed harmonics (at the frequency of the  
incoming voltage) back onto the powerlines. The problem arises when  
systems draw current from the energy storage capacitors in the ac-dc  
front-end supply in narrow, high-amplitude pulses.  These harmonics  
can cause real problems for the utility, because they cannot  
compensate for the harmonic current by adding capacitors or  
inductors, as they could for the reactive power drawn by a linear  
load.  (This is more of a concern for large industrial loads, such  
as.... oh, large water well pumps.  Not that I ever played with those  
for a living.)

These pulses contain harmonics that interfere with other equipment on  
the line and reduce the maximum power that can be drawn. In addition,  
the distorted line voltage causes capacitor overheating, dielectric  
stress, and overvoltages in insulation.

The normal solution here is to put a filter that passes current only  
at the frequency of the voltage (e.g. 50 or 60 Hz). This filter kills  
the harmonic current, which means that the non-linear device now  
looks like a linear load. At this point the power factor can be  
brought to near unity, using capacitors or inductors as required.  
This filter requires large-value high-current inductors, however,  
which are bulky and expensive.

The modern solution is to use an active Power Factor Correction (PFC)  
circuit.  A PFC circuit usually consists of a number of capacitors  
that are switched by means of a contactor. These contactors are  
controlled by a regulator (digital varmetric relay), that measures  
power factor in an electrical network.

About 30% of the world's markets—including Europe, China, Japan, and  
several states in India—have required PFCs (starting in 2001) in  
switching supplies for computing gear. Though PFC hasn't been  
mandated yet in the U.S., the IEEE is at work drafting standards.   
Still, the fact that many power supplies are now sold world-wide  
(such as the power supply for your notebook), means that PFCs are  
present in some computing gear sold in the US.
Put simply?   Modern switched power supplies are more efficient.


On Dec 27, 2006, at 6:34 PM, Brian Chee wrote:

> While it doesn't pull a constant 350watts, there is a minimum that is
> constant. This is why I'm looking towards something with a much  
> smaller
> power supply that is closely matched to the total load.
> (drives+interfaces+motherboard+CPU+RAM) This is the reason why most  
> of the
> appliances go with either wall warts or inline power supplies.  
> Typically
> these types are in the 35watt range.
>
> So for instance the 250gb snap server appliance has a 60watt power  
> supply
> pulling 44watts off that under full load. I can afford to keep a  
> 60watt
> light bulb burning, but a 400watt like in my old IBM Netfinity  
> server is
> more like keeping a small hairdryer running...a bit different.
>
> The Medallion by TechSol that I was using for the PODS project  
> didn't have
> a floating point processor (I dont' think NFS or SMB needs it)

Neither NFS or SMB directly use any floating point code.   Typically  
floating point is accomplished
via emulation when it doesn't exist.  Its either that, or you have to  
*make sure* the compiler
never generates floating point.   Else, you can end up with a  
compiler that generates FP instructions
(believe it or not, its often faster to put a variable in a FP  
register and run an operation (such as increment) there).

The Medallion boards are all either ARM7 or ARM9-based CPUs.    
Typically ARM7 cores do not implement FP,
while many ARM9 cores do.   However, the Samsung S3C2410A used on  
Medallion's boards does not, so no,
you didn't use any floating point.

> and can be
> run with a 30watt wall wart and depending upon what i/o you turn  
> on, can
> drop its load down under 10watts with ethernet+IDE+LCD running. We  
> got the
> medallion down to 32mw in deep sleep, and something like 3 watts  
> with most
> of the i/o turned off and only jumped up to 3 watts when we turned the
> radio on. (VGA, serial, USB, etc all take LOTS of power)

Serial doesn't have to if you can live with 3.3V voltage levels.   If  
you need real RS-232 though, you'll pay the price
in your power budget, because "real" RS-232 uses voltages from -15V  
to -3V for 'high' and +15V to 3V for 'low'.   These
tend to require a level converter such as the nearly ubiquitous  
MAX232 (and its clones).

USB doesn't need much voltage until you're USB-host compliant, and  
then you have to supply up to 2.5W (0.5A @ 5VDC) to the device.

radio draws a lot of power too.

> Ideally what I would love to build is a medallion (or Via) based  
> system
> running something like freeNAS on a super small power supply to float
> charge a 12volt gel cell to avoid the efficiency loss in a UPS.  
> This is
> why my home security system takes so little power, same deal. My only
> hesitancy is that I may need more CPU horse power than a strongarm  
> 1110
> can provide if I want to implement iSCSI.

If you think Via is low power, you should check to see what people  
are doing with ARM/Xscale.

The SA1100 is nearly dead.  Intel EOLed it over two years ago.   The  
new part that Intel recommends is the PXA255/PXA26x.

If you want to impleemnt iSCSI, then a 80219 or IOP3xx are likely the  
'right thing'.

Of course, I proposed a product using a 80219 earlier in this thread,  
so I'm now repeating myself.

> One thing to note, if you plan on playing with virtualized load  
> balancing
> like VMWare ESX server (VMotion) you MUST be using iSCSI at a minimum
> since SMB/NFS NAS will NOT work.

Thats just a software bug.   If iSCSI can be made to work, then SMB/ 
NFS NAS can be made to work.
Its just bits on the wire.

Jim