Path: ...!eternal-september.org!feeder2.eternal-september.org!news.eternal-september.org!.POSTED!not-for-mail
From: Don Y <blockedofcourse@foo.invalid>
Newsgroups: sci.electronics.design
Subject: Re: Redundant power supplies
Date: Sun, 27 Oct 2024 17:03:23 -0700
Organization: A noiseless patient Spider
Lines: 156
Message-ID: <vfmkcd$lb42$1@dont-email.me>
References: <vfko2a$57tm$1@dont-email.me>
 <ecpshjt023eik7conlt1fm6a8ipoa1hchj@4ax.com> <vfmace$jh1o$1@dont-email.me>
 <cgdthjt023eik7conlt1fm6a8ipoa1hc53@4ax.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Injection-Date: Mon, 28 Oct 2024 01:03:26 +0100 (CET)
Injection-Info: dont-email.me; posting-host="fa7380676466dde76cb190732ebe317a";
	logging-data="699522"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX1+WPI+Ww5o9Vfp1VO/fMANM"
User-Agent: Mozilla/5.0 (Windows NT 6.1; Win64; x64; rv:102.0) Gecko/20100101
 Thunderbird/102.2.2
Cancel-Lock: sha1:460jHhajhudbPOUCiaG4QHVHG8U=
Content-Language: en-US
In-Reply-To: <cgdthjt023eik7conlt1fm6a8ipoa1hc53@4ax.com>
Bytes: 8006

On 10/27/2024 3:17 PM, legg wrote:
> On Sun, 27 Oct 2024 14:12:44 -0700, Don Y
> <blockedofcourse@foo.invalid> wrote:
> 
>> On 10/27/2024 9:28 AM, legg wrote:
>>> On Sat, 26 Oct 2024 23:54:01 -0700, Don Y
>>> <blockedofcourse@foo.invalid> wrote:
>>>
>>>> Most of my boxes have dual power supplies.  Most
>>>> don't give me an option as to how they are used/configured;
>>>> one shits the bed, the other is there to cover the load.
>>>
>>> What kind of 'boxes' are you talking about?
>>> Redundancy is seldom a feature outside the military,
>>> telecom, life-support or un-interruptible fields.
>>
>> Servers.  Most (except cheap ones) have redundant power supplies.
>> Even some of my "non-server" boxes are so equipped.
>
> Servers are telecom, if the guy who labelled it has been
> paying attention. Obviously most PC's can do the job, but
> that doesn't qualify them to do it.
> 
>>>> I have always *assumed* they were configured to SHARE the load.
>>>
>>> Sharing is not a feature required for redundancy.
>>> Failure isolation IS.
>>
>> So far, every box that I've checked, "shares" the load (as
>> indicated by draw on the mains) to within about 10%.  Unplugging
>> either supply causes the other to carry the full load (reflected
>> by the "doubling" of mains power consumption)
> 
> Nice to have; can make other features easier; but again, not
> critical to redundancy.
> 
>>>> I picked up another box that gives me the option of NOT operating
>>>> them redundantly (what the hell does the "extra" one do, just
>>>> sit around??).  And, when in the redundant configuration,
>>>> allows me to choose which is the "primary".
>>>
>>> The former doesn't make any sense
>>
>> It would "make sense" if one supply was selected to carry the
>> entire load while the other was on standby to cover the case when
>> the "primary" failed.  Though it seems like it would be easier to
>> just say "PS #1 is primary" and let the user recable the box if
>> required.
>>
>> OTOH, if #1 is toast (or, not present), then selecting it as the primary
>> would be foolhardy.  (One assumes the system would immediately switch
>> to #2)
> 
> Select schmellect. You can't account for stupidity.

So, the server manufacturer (Dell in this case, HP in others)
are just engaging in marketeering.  And, their customers haven't
yet caught on?  Having the ability to specify *a* primary power
supply will be regarded as a *feature*??

>> I just don't see why you couldn't use a simple diode switch to
>> pick a single supply (despite the fact that there are multiple
>> output voltages... you wouldn't want supply #1 to carry V3 and
>> supply #2 to carry V1 & V2).
> 
> It's not possible to claim redundancy without safe isolation in
> the event of single-point abnormals, including direct output
> short of each unit, individually. Diode or 'active' rectifier
> isolation is basic.
> An active rectifier may be disabled (open circuit), as a form
> of intentional isolation for diagnostic testing.
> Another 'simple' feature is margining or straight 'soft' disable.
> 
> You wouldn't normally use all of these features outside a
> triple redundant system, or a single unit test could bring the
> system down, if the 'second' unit had already failed.

Everything is a study in probabilities.  The chances of TWO units failing
is less likely than *one* failing.  You weight the value of the system
(resource) being accessible with the probability of some number of
failures taking it down.

>>> - the latter is a simple matter
>>> of voltage adjustment differential and degrees of slope compensation
>>> in the sharing scheme. Some digital/software guys can get carried
>>> away - offering features that don't make any particular sense,
>>> just because they can . . . . . Simple is usually better when it
>>> comes to power delivery, regulation and security.
>>
>> The "software guy" likely was told he had to control a setting
>> that the *hardware* guy had created.
> 
> Power supplies are designed and tested to spec, in and out of
> the end-use system. . . . . but they can just be thrown into
> another syatem, after the fact, without the same scrutiny.
> Just because you CAN do something, doesn't make it either
> sensible, logical or worthwhile, out of context.

So, you don't *know* why the setting is there and why altering it
affects which of the redundant power supplies carries the load
when the machine is first powered on...

>>>> This suggests one is carrying the load and the other is switched
>>>> in (even if passively) when that one fails.
>>>>
>>>> Is there any advantage to this over a "sharing" configuration?
>>>
>>> Switch-over and single-fault failure characteristics can vary
>>> between different set-ups. The aim is to provide minimal
>>> disturbance to the system when one unit fails, or when one
>>> unit has to be hot-swapped out.
>>>
>>> Hot-swapping is not automatically feasible, just because there
>>> are two power supplies in parallel. It could be that the
>>> option to select a 'primary' unit is a crude attempt to
>>> provide quasi hot-swap capability.
>>
>> The server in question is probably the most recently designed
>> (though no idea where it sits on the marketing scale of "trim level")
>>
>> I will be working with it, next, so will watch to see *if* it
>> does, indeed, draw all of its mains power from a single supply.

Verified to be the case -- unlike the other servers.

> Hot-swap hardware is easily identifiable, as they physically
> pull-out / push-in. No disassembly required.

Exactly.  My fans, power supplies, disk drives are all hot-swappable
or hot-spare (including DRAM).

> This is not a minor design consideration and is coordinated
> in the frame, connectors and other hardware to allow
> seamless power delivery throughout.
>>
>>>> And, why would I ever want to *disable* PFC?
>>>
>>> Non-sequitur. Power Factor Correction has nothing to do with
>>> redundancy. See previous comments re digital/software.
>>
>> It wasn't expected to relate.  Rather, the setting is lumped
>> together with other power-related settings in the "BIOS".
>> Should I have created another thread to ask that?  :>
>
> Power supplies don't know bios. Perhaps it's a confused acronym for
> something else.

The "BIOS"/Setup is a common interface for the administrator to specify
parameters governing the operation of the hardware.  Just like setting
the real-time clock, determining boot order, configuring RAID
controllers and volumes, update firmware, etc.  It's just a
convenient place for "settings" that affect the operation of the
device to be accessed by the user (or, through ILOM/iDRAC/etc.).

Or, we could go back to *hardwiring* options...