Hi,

I figure there's a lot of electronics gurus out there, and I need help with a quick fix for a computer system I'm fiddling with.

A PC is equipped with a 550W ATX PSU capable of delivering 12V - 20A (plenty).

It has a RAID with ten harddrives which consume a max of 1.3A during powerup and around 0.9 during idle. The disks are spun up seperately to eliminate PSU problems. This works fine, even when the drives are up in quick succession.

During activity the drives consume around 1.3A, probably with a usage spike right at the start of activity (going from idle state).

Going from idle to activity on different drives gives no problems either, neither does having all the drives running with activity, as long as they do not go from idle to activity at the same time.

If all disks go from idle to activity at the same time, some of them power off (very randomly). This will happen if I start a defragmentation run.

My guess is that the problem is that the PSU cannot quickly enough deliver the extra 4-5 (possibly more just as the drives go from idle to activity), thus making some of the drives power off as they do not get enough power.

The problem disappears if 5 drives are powered from a seperate PSU. The problem gradually disappears when less drives are connected, and worsens when more drives are connected.

Is it possible to fit a capacitor on the 12V line, giving the PSU a better chance of delivering the power that is needed?

I need to fit capacitors to two drive arrays with 5 drives, each array consuming 5A idle and around 7,5A during activity (worst case).

HOW? (Spell it out, please)

I know - "just fit another PSU", but no, I'd rather not, as the computer is stuffed, and the current PSU is capable of sustaining the load.

Feel free to email me privately, as I know this is O.T.


Thanks in advance,


Lars

You'll probably need at least 4 capacitors the size of a Saturn 1B third stage, wired in parallel to do what you are describing . The cabling would need to be pretty hefty as well...

Capacitors (for some reason) seem to have become the be-all and end-all of supply problems in this board; don't forget, they only store charge. Without a resistance in series, a capacitor can deliver all it's charge in one burst (current is flow of charge). So adding capacitors to your PSU will make other problems, like the inrush current required to fill them when you turn it on, like the potential shock risk when the system is turned off, and they still won't give you much help with your problem.

By sticking your RAID array inside your CPU, you are at risk of causing the problem the array is designed (in one aspect anyway) to address - CPU power failure. If your main power supply goes out, your RAID array should not be affected and at least store any data saves committed before the failure. Stick 'em in the same box and you'll loose the lot if the single PSU goes down.

Cheapest, sanest solution - buy yourself another CPU casing with a high power PSU that has a peak rating above the maximum current draw of all the disks at active state. Sling a cable across from your CPU RAID controller, and Bob's yer proverbial.

Doesn't cost much, reduces the risk of SPOF failures, and makes my blood pressure go down a bit

One of the few remaining Mk1 owners... #00015

I know - "just fit another PSU", but no, I'd rather not, as the computer is stuffed, and the current PSU is capable of sustaining the load.

There's a couple of problems I see here. First, you're assuming the existing PSU is capable of sustaining the load. But you just got done telling us how your diagnostics determined that it can't, in fact, handle the load. Listen to what you're saying.

The other, larger problem I see is this: You're trying to stuff ten hard disks into a PC's casing. That'll work for about three months before the combined heat from the drives begins to cause them to break down, one at a time. Drives configured in stacked arrays like that need lots of ventilation. Dedicated fans directly blowing on the drives are essential in that kind of a system.

I've had thermal failure troubles when stacking as few as three hard disks, and you're doing ten.

I'm pretty sure they make external cases specifically designed for the sort of thing you're trying to do. Research it. If you don't, the failures and maintenance will kill you in the long run.

___________
Tony Fabris

You're trying to stuff ten hard disks into a PC's casing.
I am under th impression that he isn't using a standard PC-case; this sounds more like a HP-netserver with hot-swappable diskbays.
I would also look at other causes : is the SCSI-bus properly terminated. Are you using single-ended or differential SCSI? Are you're drives single-ended or differential? Is your termination single-ended or differential? I've had random SCSI-bus resets caused by slow tape-drives which were connected at the same end as fast SCSI-disks. Maybe it's just one faulty disk causing this behaviour.
Just a thought...

Frank van Gestel

Hi,

Thanks for all your input. Everyone I suspected that would be quick to respond did so :)

As I myself work with computers, I know to distrust everything I hear from "users". Unfortunately I'm not a user, but that probably doesn't make me more trustworthy :)

Tony and Rob, I *did* listen to myself, and I do have my reasons to want to do things the way I described.

First thing, the RAID isn't for safety measures. This is a workstation. The disks are RAID0, purely for performance. Everything stored is either temporary or backed up.

Frank is right, this is no card-board box. The casing is an Intel SC5000 server casing, designed to hold the 10 drives. It has enhanced cooling, with a total of 7 fans. The case reaches around 30C when running for longish periods. The drives specifications state that the drives maximum ambient temperature to meet MTBF specs are 45C. The air is sucked in from room temperature, passes the drives and exits into the cabinet (which causes the temperature to rise from the 22C it is in the room to the around 30C the air is when it exits the casing).

Also my diagnosis does not tell me that the PSU cannot sustain the power load. The PSU has been bought as what I thought was with enough headroom. It is designed to deliver 20A at 12V. The drives use 13A at 12V when they are stressed.

My diagnosis tells me that the PSU cannot quickly deliver the increase in amps that all the drives need for a very short period of time, if the demand is all drives at once. It is the going from idle to activity state that causes the problem.

There is no problem running all disks at full activity for hours from the single 550W PSU, as long as they do not all *start* seeking/reading/writing at the excact same time. There is no problem running them all at the same time (doing hefty disk access).

Frank, it isn't SCSI resets. I thought so as well, but the drives physically power off. If it was anything but power, why does it run so well with an extra PSU?

Termination, SCSI-controller, cabling, enclosure firmware has been checked. Everything flies perfectly with an extra PSU.

I do think I did my homework before asking here....


Best regards,


Lars

The disks are RAID0, purely for performance.
... as long as they do not all *start* seeking/reading/writing at the excact same time
How are your stripe-sets defined? 10 disks in on stripe-set should also cause the effect mentioned, as all data gets distributed across all disks.

Frank van Gestel

Okay, good to know you've got the cooling thing handled.

Sounds like your only options are a beefier power supply or a second power supply. Agreed that the one you initially purchased should handle the load, but your empirical evidence suggests that it can't handle the load under all possible conditions. Rube Goldberg solutions with capacitors are the wrong way to go about it; get the right power supply for the job.

___________
Tony Fabris

Frank,

It's one large stripe set, which gives excactly this problem if I start a defragmentation run.


Lars

Also my diagnosis does not tell me that the PSU cannot sustain the power load. The PSU has been bought as what I thought was with enough headroom. It is designed to deliver 20A at 12V. The drives use 13A at 12V when they are stressed.

Two points occur to me:

(1) The PSU may be designed to give 20A on the 12V rail, but that doesn't absolutely guarantee it does.

(2) The drives take 13A. Are you taking into account anything else that might also be running off 12V? This rail is plumbed through to the PCI/ISA/whatever else slots as well, and some cards do use it. Also, what supply do the 7 fans use?

Patrick

Opinions expressed in this email may contain up to 42% water by weight, and are mine. All mine.

The casing is an Intel SC5000 server casing

Are you sure this:

http://support.intel.com/support/motherboards/server/ta_347.htm
http://support.intel.com/support/motherboards/server/ta-351.htm

isn't happening?

Peter

PS. that "workstation" is substantially better specified than the compute server used by the whole of empeg (which is also in a SC5000)...

Peter,

The firmware on the backplanes have been upgraded accordingly one week ago. So this shouldn't be the case. But thanks anyway.


Lars

Hello,

After consideration, how much I may loathe to agree, the PSU is probably not powerfull enough.

Would it be possible to connect a smaller PSU in parallel on the +12V and +5V side?

Perhaps one of the smaller PSU's from an external SCSI-enclosure or something like that. That could give the few extra amps needed.


Lars