Selfmade redundant PSU using 2x ATX?

[^rooker]

I'm evaluating the possibility of combining standard, hi-watts ATX PSUs to form a redundant power supply to drive a harddisk storage server with 45 HDDs.

The reason for that is, that all "professional" storage/server chassis for such a use-case have proprietary, non-standard form-factors (and pinouts) of their PSU modules. I'm looking for a power supply solution which can be kept alive by using standard, replaceable, documented hardware.

How much effort could that be?

The design is based on version 2 of the storage pod design by "Backblaze". In their setup, they have enough pods to provide the required redundancy in case some PSUs bail out - so for them non-redundant PSUs are sufficient.

I've already started to do some reading:
DIY redundant psu... (hardforum.com)
Power Supply System Integration Part 1: Single Source Fault Tolerant Power Systems (powerelectronics.com)
ORing diodes (codemsys.com)

When I found this forum and saw that the site was named "DIY SMPS", I knew I just had to ask here before I make any move

Thank you very much in advance,
^Rooker

 

[MicrosiM]

Hello Rooker.

I was reading about the ide.

But if I understand what you need, is connecting multiple (Standard ATX) power supplies in order to get a redundant power supply system?

in other meaning, paralleling the ATX power supply ?


Just let me know more info please

 

[^rooker]

Yes: parallelizing the ATX power supply.

By default, the Backblaze comes with 2x ATX PSUs (Zippy PSM-5760V, 760W). So my assumption is to use 4 of these PSUs and parallelize them 2x2.

My initial expectation was to just use redundant PSUs with ATX form factor, but their backside does not match the default ATX layout and can therefore not be mounted in e.g. a Backblaze chassis

Since I don't have enough space in the chassis for 4x PSUs, I'd have to have them mounted externally anyway.
Therefore, if I can't parallelize 2x ATX as a redundant PSU, "Plan B" is to buy redundant PSUs with ATX form factor (like the "MRG-5800V4V") and have them externally.
This would still be better than some completely proprietary form factor (regarding sustainability and replacement parts), but again, there's no standard for the power-modules per PSU - even if their external form factor has ATX size...

 

[MicrosiM]

Yes: parallelizing the ATX power supply.

By default, the Backblaze comes with 2x ATX PSUs (Zippy PSM-5760V, 760W). So my assumption is to use 4 of these PSUs and parallelize them 2x2.

My initial expectation was to just use redundant PSUs with ATX form factor, but their backside does not match the default ATX layout and can therefore not be mounted in e.g. a Backblaze chassis

Since I don't have enough space in the chassis for 4x PSUs, I'd have to have them mounted externally anyway.
Therefore, if I can't parallelize 2x ATX as a redundant PSU, "Plan B" is to buy redundant PSUs with ATX form factor (like the "MRG-5800V4V") and have them externally.
This would still be better than some completely proprietary form factor (regarding sustainability and replacement parts), but again, there's no standard for the power-modules per PSU - even if their external form factor has ATX size...

Did you try to connect two ATX power supplies in parallel?

and testing them?

I think you should make an external test first, and then you will be able to decide if the idea will fit with your application or not.

I think you should be able to parallel these SMPS units, with modification maybe, I never tested.

If you have two units, lets start some testing

 

[^rooker]

That was my very first idea, actually: Just hooking 2 ATX PSUs in parallel.
But then I read up on redundant PSUs and found out about the "ORing diodes", which led to "ORing MOSFETs" - and since electronic components exist for just that purpose, I assume there's a reason for it.
And electrically I do understand it (a bit) now.

For example:
Texas Instruments TPS2410
Maxim MAX8535/8536/8585
Intersil ISL6144

This storage server must be production stable, and so I need to have really reliable redundancy of the PSU (avoiding power-spikes during fail, etc...)
I do have older PSUs for "playing around" for a prototype.

 

[MicrosiM]

That was my very first idea, actually: Just hooking 2 ATX PSUs in parallel.
But then I read up on redundant PSUs and found out about the "ORing diodes", which led to "ORing MOSFETs" - and since electronic components exist for just that purpose, I assume there's a reason for it.
And electrically I do understand it (a bit) now.

For example:
Texas Instruments TPS2410
Maxim MAX8535/8536/8585
Intersil ISL6144

This storage server must be production stable, and so I need to have really reliable redundancy of the PSU (avoiding power-spikes during fail, etc...)
I do have older PSUs for "playing around" for a prototype.

The only reason i told you make some protos, because I cannot guarantee that it will work, you may need some thing to solve the voltage diffidence problems

 

[^rooker]

I was definitely planning to build prototypes, because the final PSU must work reliably.
There seem to be more issues here than I initially considered (because I didn't know better).
These are:

1) More than just 1 cable per voltage:
I knew that ORing had to be done per voltage (+5, +12, ...), but I didn't know that power supplies had multiple internal circuit sources even for the same voltage.
So, for example: ORing +12V might require multiple ORing MOSFET ICs, depending on how the PSU model is built internally. This makes it quite difficult to build an ORing adaptor where you could simply connect 2 ATX PSUs - regardless of their model/vendor.

2) "There is more to power than just Watts."
It's great that Backblaze is publishing so many of the knowledge they've gathered building their pods. On the page of version 3 of their pod-design, they explicitely mention that:

Each vendor imposes unique limits on the amount of power you can draw off of each rail and unused power on one rail cannot be used on another.

This means, that the assumed gain of building it with standard ATX PSUs does not necessarily mean that you can use any other PSU - even if the supplied power (watts) match.

3) Free form factor:
The biggest problem with proprietary PSU modules in "professional" server/storage equipment is, that you have absolutely no option to replace it with anything else - due to the physical form factor restrictions.
I was hoping to eliminate that restriction by allowing the PSU to be outside of the actual chassis.
But I also assumed that I'd only have to supply 1 cable connection per-voltage. As mentioned in (1), this is not the case (if I understood it correctly)


== Conclusion:
If it would be possible to design a generic ORing circuit that allows using (almost) arbitrary ATX PSU models in order to behave like a redundant PSU, the effort might be worth it.

With the above considerations, and given the current state-of-the-art situation, I'm not so sure anymore if this is possible...

 

[^rooker]

Oh, one more thing:
Actually, I'm pretty mad at the vendors for not agreeing on a standard form-factor and pinout for their PSU modules.
I mean, if you take a look at Supermicro's PSU matrix, it's clear that there are just too many (and that's just one single vendor).

This makes it hard to keep hardware in use for a longer period than the vendor's planning their product life-cycle.

 

[norazmi]

i think you can use multiple computer psu without parallel them but you can have them run at the same time, let say one psu run 5-6 hdd, you can make one switch to power on them all by using remote cable green and short to black cable (ground). because in my experience computer psu have protection against over current, overload, over voltage and under voltage, and i did experiment with psu using tl494 switching when i remove resistor ocp and diode 1n4148 and yes u can paralalel them but it will run without protection i think it will not good for long term usage.

so the solution is run them without parallel but lets them all power on together and its about the same usage of parallel them and safety .