2A/3A fuse to discriminate with 16A b-curve MCB?

[very_pointy_ears]

Hi,

I'm looking for a fuse (unless someone comes up with a better idea) to power a computer that, in the event of a short circuit fault in the power supply, will blow while not causing the 16A b-curve breaker that feeds the curcuit to also trip. The circuit is already drawing 12 - 16 amps constantly.

In more detail:

At work we rent power and rack space in a few data centres, and have a few dozen 19" racks of computer equipment (1U servers, mostly). Each rack has its own single power feed via a 16A b-curve breaker back at the data centre's power distribution board. We're not in a position to change the breaker because it belongs to the data centre, not us.

Each rack may have 20 - 30 servers in it, each drawing several hundred milliamps during normal operation. Total draw per rack is approx 12 - 16 amps.

On several occasions we've had power supply failures in a single server that has caused a momentary short sufficient to trip the breaker supplying power to the entire rack, causing all of the equipment in the rack to shut down.

Power distribution *within* each rack is down to us. Currently we have some remotely manageable power distribution units - one 13A plug supplies power to a PDU, which has 8 unfused IEC outlets. We have three PDUs per rack and these feed power to most of the equipment.

I was thinking of the possibility of having an IEC lead with an inline fuse holder with maybe a 2A fast blow fuse in it, maybe a 2A variant of this:

http://www.farnell.com/datasheets/10867.pdf

Am I barking up the wrong tree, or would a 2A variant of the above discriminate with a 16A b-curve breaker that's already drawing close to its rated load?

Thanks for reading through this far

Cheers,
Kingsley.

 

[TicklyT]

A problem you may have is the initial inrush current of SMPSs when initially powered up.
They will often wipe out a cartridge fuse rated at several times their rated input current.

The equipment I used to work on had BS1362 7 Amp fuses in plug tops supplying 90 Watt SMPSs, as 3 or 5 Amp fuses would nearly always pop when they were powered up. The 2 Amp time delay fuses within the power supplies always survived.

 

[Chri5]

Surely UPS would be prudent

 

[very_pointy_ears]

Surely UPS would be prudent

I presume you're suggesting a UPS in order to absorb current surges such as this, rather than expecting it to continue to power the entire rack during the night if the breaker trips at, say, 3am?

Have to admit I hadn't considered using a UPS for this, so good shout. I don't know about this particular data centre but the other one (which we've used for longer) has 2 separate connections to the national grid plus its own backup generator and so we've never considered a UPS a requirement there (esp as they have 20A c-curve breakers which we've never managed to trip).

 

[holmslaw]

..

 

[ban-all-sheds]

Your last point shows what the root of the problem really is - the Mickey Mouse datacentre - those racks should be on at least 2 feeds from two different phases, and all the servers should have dual redundant PSUs.

(Actually - racks full of 1U servers should be changed to blades, but that's another matter... )

 

[very_pointy_ears]

Thats a lot of servers in one rack, I would not expect to see more than 10ish per rack, with an empty space between servers.

The amount the data centres charge for rack space makes it unviable to put only 10 or so servers in a rack. Customers wouldn't pay the resultant price to colocate or lease them, unfortunately, they'd just go to someone else cheaper.

Have you actually recorded the current, because your figures seem low to me.

It does vary depending on the server spec. We try to stick to servers with a lower power footprint wherever sensibly possible (eg LV processors and things) although /some/ servers draw almost an amp. The data centre staff pointed out that, upon restoring power to the dead rack this morning, draw was 14 amps for the whole rack.

Server racks such as yours can get very hot due to lack of air circulation and this will increase the current consumption. The server manufacturers advise on this can be ignored because it will be based on some kind of perfect/ficticiuos air conditioning system.

Agreed, we've had these problems in the past with some of our older glass fronted racks, despite fitting fan trays to improve air flow. The fans then run at full speed to try to keep the server cool and these draw noticeably more power, thus compounding the problem. The racks in this data centre have fully vented front and back doors with a hot and cold aisle. Cooling seems to work reasonably well.

Are the racks designed for upflow or downflow this will obviously effect the cooling.

Front to back flow through the vented doors.

Whats a "momentary short"? is there a zinc whisker problem [google if you don't understand] there is also the possibility that metal shards have been added to the screed, these could similar problems.

I don't know how long these shorts last for, unfortunately, as the first we know about it is that we get an outage and when someone goes to look at it, the rack is powered off and one of the servers doesn't come back on when re-presented with power, and it turns out to have fried its PSU.

I do know what you mean by whiskers, at least I think I do, without resorting to google - I've heard of metal whiskers growing over time (a bit like a crystal does) and causing electrical problems. I'm not sure I remember what causes them though.

Of all of the hardware types we have in our racks, this problem has only happened with one particular manufacturer. Unfortunately, although we're wise to the problems these days and don't sell these any more, quite a few legacy kit remains

 

[very_pointy_ears]

Your last point shows what the root of the problem really is - the Mickey Mouse datacentre - those racks should be on at least 2 feeds from two different phases, and all the servers should have dual redundant PSUs.

Different phases in the same rack?

I thought that doing that would have been considered a generally bad idea, though I'm happy to be proven wrong.

(Actually - racks full of 1U servers should be changed to blades, but that's another matter... )

We've looked at blades but not really considered them much at this stage. They don't help when a customer needs to plug in kit that they already own themselves and just want power and connectivity, though, as opposed to leasing the kit from us (in which case we could sell them a blade, or a virtualised server, etc). But I agree, that's another matter

 

[ColJack]

no, 2 single phase supplies from different phases.. that way if a phase goes down, you can swap the plug over to the other phase and get back up pretty fast..

 

[AndyPRK]

well back to the question.......


I think you need to measure the startup current to find out the correct fuse to use.


I think it is a good cheap solution and worth trying to see if it does the trick

 

[holmslaw]

..

 

[plugwash]

Getting back to the original problem how about a small CU in the rack filled with B6 breakers. Then wire your servers power cords into those breakers.

Failing that could you replace the PSUs in the servers with a more reliable brand?

 

[JohnD]

not my subject but, a place I used to work had several DIN rails in each cabinet; IIRC each had an MCB (lets say 6A) feeding a row of (say 6a) fast-blow fuseholders, one per device. So any one device would blow its own fuse faster than the MCB would trip. The fuses were rated high enough not to blow from a start up or other planned surge.

 

[ban-all-sheds]

In every data centre a decision has to made as to the crossover from power resilience to computer resilience. EG. google use multiple bog standard interlinked computer systems (I believe five at least per system) located around the world to maintain their business, the alternative would be invest in one or two mega expensive data centres.

Google are unusual in that their servers (at least for the search services) can all be stateless.

At the rack level you would not have duplicate feeds, and none of the kit would be fed from different phases, they could be fed from different sources but not different phases.

Yes you would and yes you do and yes they can be and yes they are.

Not correct, see above for crossover from power to computer resilience.

I bow to your superior knowledge of Mickey Mouse servers.

If the wink is meant to imply some secret knowledge, it does not.

No, it's not secret knowledge, lots of people have it.

But it does demonstrate your lack of practical application. Installing blades will make the ops problems worse due to the higher cooling requirements.

The cooling requirements per blade are not higher.

True, if you stuff the rack full of blades then the rack will need more power and more cooling, but if you put in the blade equivalent of a rack-full of pizza boxes you'll use less. Don't forget that by using blades you also take out network & SAN switches.

 

[holmslaw]

..