Power Supplies and Multiple Motherboards

[Quartzeye]

Is there a way of hooking an unknown number of power supplies to an unknown number of motherboards.

I was thinking that there must be a way of modifing the power supplies to feed power to a "Power Buss" of some kind. Then I think you could plug in a number of motherboards into the "Power Buss".

Is this possible? Isn't this how hot-swappable power supplies work?

What I want to do is to build a custom box to hold five motherboards that make up a web server cluster. I don't want to put five poweer supplies into the box. I was thinking I could get by with two or three 600watt power supplies. I don't plan on OC'ing the boards but I plan on water cooling the contraption.

Drop me a line if you all have any suggestions.

Thanks in advance.

[resago]

let me know if you need any of the atx PS connectors, I have several from non-functioning PSs.

[Brad]

yeah, just connect the 20pin connectors all up together from the PSU's, lets say 3x 20 wires going into a board, and say 5x 20pin outlets going to the mobo's. it'd work fine

[futRtrubL]

I would feel better with some diodes in there to stop shorts in case of voltage levels differing between PSUs, but the voltage drop wouldn't be acceptable.

Edward

[Doraemon]

Anyone know where I can buy the power connectors to make such cable?

thanks.

[bigben2k]

I'm not trying to knock your idea (it's fine) but have you considered rackmount?

[Brad]

kdcomp have these: http://www.kdcomputers.com/eui/prime.../component.htm

I'd cut these in half and use the female side to connect psu's into the board, and the male end to connect into the motherboard cable

[Quartzeye]

Thanks for the info!

Since ATX power supplies are switching, I was under the impression that you could not just wire up the connectors from multiple supplies. Am I confussed????

The rack-mount idea is what I am trying to accomplish in function but not in form exactly.

My idea is to take 4 - 8 m/b's, cheap ones with built in video, and set up a huge cluster.

I plan on building a case that would house the m/b's in a side-by side fashion. I wanted to plug the m/b's into a bus that was powered by 3, 500 watt power supplies. Not just for the power but for some redundency.

Each m/b would would be plugged into a "water-buss" so that I could keep everything cool and drop the noise level.

As this would be a server, I planned on building a network switch and KVM into the case. That way I would have access to the cluster as a single unit throught the cluster controller board.

Running dual nics on each board was an issue until I found 90 degree riser cards.

I figure with 8 m/b's I can get a whole lot of power in a 16 X 16 X 20 custom case.

Maybe, what I want to do is the wrong approach, but space and cost are an issue when compared to rack-mounted systems. Not to mention space and noise as well.

Any additional info, schemaics, etc. on wiring up the p/s's and switchs would be greatly appreciated.

[bigben2k]

It sounds like you're trying to somewhat reproduce Hewlett Packards Blade Server idea. RLX and IBM also make some.

There are ways to do the power thing, and for redundancy, you'd need to start thinking about hot-swap n+1 supplies: have enough PSUs to power everything, and then add another one. Set them all up so that you can swap one of them, live.

It's possible to do it, but you're going to get into some pretty heavy wiring (barbuss). You'll also have to tune the PSUs so that they provide as close a voltage to each other as is possible. See pHaestus' article on this site.

If you're considering dual NICs, then I have to assume that you've already considered dual CPU mobos, right? And then server clustering?

Where did you source the PCI risers?

What on earth would you need so much processing power for? A home-made ISP?

[jtroutma]

Even a large ISP doesnt require THAT much CPU power..... were talking minature rendering farm here

[Quartzeye]

You nailed it. I want to create a blade-like server but use standard off-the-shelf hardware. And of course clustering is required for my purposes.

I definately want to build a web-server cluster. The power comes from the increased response time I'll have due to the cpu availability. I read an article over on a linux site about cluster web-servers and the metrics they had indicated that you really pump up your ability handle a large load on your site.

I did look at the ECS dual Pentium boards but they cost more than some budget P-4 boards. I am sitting on 5 socket 428 p-4 boards that I am planning on using in the cluster. Got them at a price that was to low to pass up. Anyway they should be fine for what I am trying to do.

Can anyone explain the "barr buss" or how to balance the power supplies? Also, what makes a P/S "hot-swappable"? It has to be more than standardized connector. Heck I can buy something that will do that and create a removeable case kind of like the removeable hard-drive kits that sell for @$10.

I have to believe there is more to it than that. What happens if I have two+ P/S's plugged together but one is not switched on? Will I damage the P/S? What about plugging and un-plugging one of the P/S's? Lastly, is there a problem with turning on the P/S's one at a time? Do I need to wire them up to all trip on from the same switch?

I know it's a lot of questions and I really appreciate all the help sent my way.

Thanks.

[bigben2k]

The bar buss is simply copper bars mounted on an isolated board/panel, where each bar carries a different voltage.

since you'll have 5 dual mobos, you're probably looking at 2500 watts of power (probably less), and on the 5 and 12 volt lines, that's a lot of amps. Now you can use heavy copper wires, if you want, but the busbarr MAY be simpler.

I'd suggest that you read pHaestus' article on using multiple PSUs. Check the main site's page.

A PSU is hot-swappable if you can remove it while everything else stays on. Keep in mind that if you remove one, you still have to have enough power to keep the system running, hence adding an extra one. The connectors are up to you: just make sure you use something that is easy to deal with, and that it can handle the power load.

IBM has a beautiful feature on some of their servers: hot swappable PCI cards: press a button to turn off the card, then remove it. The idea here is similar.

To balance the PSUs, first stick to the same model. Turn them on (see pH's article) then start measuring the voltages. You can tweak those voltages to match.

Alternatively, if you want to get away from doing that, then add 2 extra PSUs to your power subsystem. That way, if the load isn't balanced, at least none of the PSUs will end up running over specs.

Good luck.