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

Re: (OT) Storage (8*IDE HDs) any experiences?


On Tue, May 01, 2001 at 07:46:42PM +1000, Hamish Moffatt wrote:
> On Tue, May 01, 2001 at 11:03:06AM +0200, Russell Coker wrote:
> > On Monday 30 April 2001 00:04, Hamish Moffatt wrote:
> > > I don't see why. Nor is this any different to any external drives.
> > > You have a hefty ground connection between the power supplies anyway
> > > (the mains, plus the metal case acting as ground).

Alas, one generally cannot assume this!

Here in The Netherlands, most residential wall outlets do not have a
ground connection.  This causes mysterious computer illnesses.

The server won't come up on reboot, when the printer is attached.
The problem is gone after the printer gets a proper earth ground.
After purchase of a new "Brand Quality" monitor, the harddisk appears
damaged.  The bootup ID string is corrupt after 12 or so characters.
Fixing the monitor's wall outlet fixes the harddisk.  Multiply the fun
as you start plugging in network cables.

But I digress..

Frame ground isn't so important here I think, circuit ground is what you
should care about.  In principle, these are different things.  In most
applications they also are strictly different things, but pc's just
isn't one of them.

> > External drives generally don't use an ATA interface!  I am not confidant of 
> > the main earth acting as a suitable earth for the DC power.
> True, but I don't see this as a big issue.

All traditional external drive systems that I know of have:
- Their own power supply in the external disk housing.
- A connection to the host that carries only signals, not power.
  This connection has a dedicated frame ground line (in fact,
  any decent cabling has its shielding connected to frame ground.)

> I think it would be better to deliberately turn them on in order, rather
> than trying to guess at the same time. Turn the hard drives on first.
> They may or may not spin up while the controller is powered off. Then
> turn on the main supply.

AFAIK harddisks have two motors: 
- A start motor that speed up the disk's rotation very quickly, but eats a 
  lot of current;  It is normally only used at boot time.  
- A continuous motor that spins with a very precisely controlled speed and 
  consumes considerably less power.  It works all the time when the disk 
  is operating normally.

If some day you turn on your computer and suddenly the disk is dead,
you should be able to hear from the disk's cries of agony which of the
two motors burned out ;-)  (There is an (urban?) legend that harddisk
manufacturers classify batches by the motor that is expected to fail
first.  If the start motor is weak, it will be a scsi disk, if the
continuous motor is weak, it will be an ide disk.)

All scsi disks (that I know of) have this feature, called "Spin Delay".
If you configure the disk appropriately, it will not attempt to spin up
on powerup until it is explicitly asked to do so when initialised by the
scsi host controller.  This way the system can distribute the surges in
current draw caused by the powerful start motors in the disks.

> > There was a presentation at a Linux Users of Victoria meeting some years ago 
> > about doing hot-swap IDE hard drives with cheap standard hardware.  My 
> > recollection is that the power lines of the hard drive had to be connected in 
> > a particular order...

I have sucessfully powered down, disconnected, reconnected and powered
back up again an IDE disk once (this is why you should take anything I
claim here with a grain of salt.)  No umount or even swapoff, just disable
dma and cross my fingers ;-)  The disk was only off for 30 secs or so.

What probably helped is that the disk is an old, low-rpm disk.  Modern
disks seem to have a tendency to draw a lot power at once when power
is plugged in from a running system, throwing the whole system into a
hardware reset.  A rather unfortunate side-effect when hotplugging.

It can be really nice to have cheap (free) old hardware (junk) to mess
around with.  How else would I have discovered that sometimes, you _can_
successfully hotplug isa cards  ;-)

> Standard power supplies may have sequencing to switch the supplies on a
> known order. That doesn't stop you powering them from different power
> supplies though, as the sequencing isn't under motherboard control.
> > > On Monday 30 April 2001 16:11, PiotR wrote: 
> > > A good solution for this might be to connect the first PS's output to the 
> > > other, so the voltage is the same, and there's no massive current flow 
> > > across the data cables.  
> > 
> > That's if both PSU's have exactly the same voltage.  If one provides a 
> > slightly higher voltage than the other then it will try to power everything 
> > itself (at least until the current drain lowers the output voltage).  Also if 
> > two PSUs with different voltages are connected together with insufficient 
> > load then reverse current will flow through the PSU with the lower voltage!

Well, I'm not an electrical engineer, but I don't think it really works
like that.

   *** WARNING ***
My understanding of switched-mode psu's is limited and I am not an
electrical engineer.  So don't listen to me.  Even worse, I learnt some of
what I claim to know on the USENET.  I'm dangerous!  All your insurance
policies will be void if the insurance companies find out that you have
been reading this message ;-)
   *** WARNING ***

AFAIK It is very hard to get a reverse current to flow through a
switched-mode psu.  You would have to exceed the reverse breakdown
voltage of its internal rectifier diodes in order to do so.

If one sm-psu gives off a little less voltage, it will transfer an
accordingly little less power to the shared load.  It will not stop
pumping current altogether, just (proportionally) less.  The only negative
current I see would be the one that expresses difference with the expected
current in an ideal 50-50 split between two psu's.

Actually I'm told that, by their method of operation,  switched-mode
psu's are in principle very well suited for parallel configurations,
contrary to the popular lm78xx type of _linear_ regulator circuits.

But you probably do need to worry about at least three issues (and here
I already assume that you have given up worrying about your life when
you try any of this):

- Failure modes:
If the psu fails "open" on the secondary side, then that is good, it just
means that the other psu must work twice as hard (but it might also
overheat as well in 5 secs.)  Adding more parallel supplies makes sense
when dealing with open failures.

If the psu fails "closed" on the secondary side, then the other psu would
see a short circuit and if you're lucky would just shut down gracefully.
Adding more parallel psu's doesn't help anything but start a bigger fire.

- Oscillations:
May be a particularly tricky problem to deal with.   When you add two
psu's, the internal feedback circuitry in each of the psu's now gets
feedback from the other psu's output as well.  This could cause the
combined circuit to be more prone to unwanted resonation or "ringing".

Add more psu's and it probably gets worse.  All you can do against this
is to modify part of the feedback circuit in the output stage of all of
the sm-psu's to take the feedback from the other units into account.

At this point it is probably better _and_ easier to design your
own fully-integrated multiple-redundant switched-mode power supply
circuit and use the standard pc pcu's as a cheap source of components.
Don't forget to count your hours spent on it ;-)

- Consumption imbalance:
Harddisks commonly use +5V and +12V power inputs.  Motherboards also use
a lot of power at a +3.3V level.  Standard ATX power supplies are designed
to provide these different outputs with a minimum consumption on each.

Consider what happens if you build a monster system with 20 harddisks,
one mainboard and 5 load-balancing/redundant psu's.  Then each of these
psu's provides 1/5th of the amount of power consumed by the mainboard
at +3.3V, which may well be out of spec for proper normal operation of
the psu.  It might shutdown or automatically go into some powersave mode
that also cuts off 95% of +5V and +12V power to the disks.  It might
appear to continue to work, but be much more sensitive to oscillation.

IIRC most pc power supply units really regulate only one output line and
keep the other outputs at a fixed multiple of the carefully regulated
output.  If specifically that output is underused, a breakdown in either
the basic regulation mechanism or in the fixed rate mechanism can be
the result.  In practise, the +12V could suddenly be +15V, or maybe +35V
in transients that do not show up on an average volmeter.

In conclusion,

If you're still with me, you probably need to get a real life.  It might
be a very simple hack onto a standard made-in-taiwan pc psu, provided that
you really have acquired a thorough understanding of sm-psu circuit design
and all of its associated peculiarities (better get a life instead).

Unfortunately, that takes a little more knowledge than Joe Random Nut's
potpourri of warmed up random facts that he found on the usenet, the
weekly "science" section of the friendly local newspaper, and the
instruction booklet/folder/paperscrap that came with the soldering iron.

In other words, if you have to do this, do it at home or any other
place where people can't see you doing this sort of crazy stuff and most
important: where they can't get hurt from it innocently.

> Yes, that's a bad idea.



Reply to: