Re: Redundant power supplies

On 10/30/2024 5:54 AM, legg wrote:

On Tue, 29 Oct 2024 23:33:49 -0700, Don Y
<blockedofcourse@foo.invalid> wrote:
 

On 10/27/2024 8:00 PM, Don Y wrote:

Apparently the settings have nothing (little?) to do
with reliability/redundancy.  Rather, they are there
to improve energy efficiency (!)
>
  From the long list of settings he sent me, these people REALLY
try to save every watt they can!  I guess if you have
thousands of servers, a few watts on each has consequences
(cooling, etc.).
>
[As a quick test, I was able to change the power requirements
for my server addressing a fixed load by more than 10%!]

>
After playing with assorted setting combinations, it
is obvious that they do nothing to affect the (shortterm)
availability of the server.
>
When configured to share the load, the load is approximately
split evenly between the two supplies -- if both are powered on.
Unplug either and the other takes the full load.
>
When configured as hot spare, select PS#1 as "primary", unplug
that power cord and PS#2 "spins up" to take on the load.
>
Unplug PS#2 and, as expected, no change (other than a
warning indicating that you are now completely reliant on
that ONE power supply).
>
*BUT*, the total power consumed goes UP when the load
is shared if the power supplies are "lightly" loaded.
This makes sense as efficiency tends to improve with
the magnitude of the load.
>
In my case, there was about a 25W "penalty" for operating
in the sharing configuration (on a ~220W load -- single CPU,
external SAS HBA, 8x1TB, 256GB DRAM).

 Nothing is free, but you will have double the hold-up time.

That depends on how much is handled "inboard" of the power supplies.
Obviously, the power system has been designed to support the rated
load of the individual power supplies *during* such a hand-over.
And, regardless of mains voltage (autoranging) or power line
frequency.

Moral of story is to size individual power supply to handle
the entire (projected/future) load and then use the hot spare
to give you redundancy.  So, this is an enhanced feature that
isn't available in my other servers...
>
[Amusing as I tend to oversize the power supplies to address
unknown future needs... annoying to have to upgrade a power
supply -- PAIR of power supplies -- just because you want to
embelish the hardware in a box!]
>
But, still no idea why I should be able to *select* the primary
supply if both are cabled!

 Not all redundant supplies are hot swappable - but rely on the
supply being pulled/inserted to be in a disabled/dead or powered
down state.

Not the case, here.  These are marketed as hot-swappable.

It's a connector pin sequencing and dynamic start-up/shutdown/
engagement performance issue.

GND pins make first and break last.
Additionally, the system involved (memory, power, storage)
has protocols to make this a graceful transition and indicators
to convey the state of that protocol.  I have laminated
cheat-sheets (business card sized) for each host to know
how to affect those changes on that particular host as
there is no "standard".
E.g., the power supplies indicate when they can be swapped
out via an indicator on the supply.  They indicate when
they *should* be swapped out (mismatched, etc.) using the
same indicator.  The log and front panel provide more
human readable indications.
Storage devices that are part of a managed array will
typically idle themselves when failed.  Or, indicate
when they are being rebuilt.  SATA/SAS devices can be
treated as hot-swappable if you umount() them to ensure they
are quiescent.
[I suspect MS "server" OS's have this covered in the
Management Console.  In the *BSDs, I've found you
have to go through several explicit steps to unmount,
remove, reattach, reprobe, remount replacement volumes)]
But, even subsystems that don't explicitly support hot-swap
can be coerced to do so.  You just can't do so "willy-nilly"
and expect the device not to incur a loss.