Another dead power supply

[Peter]

On Tue, 6 Dec 2005, John Macdonald wrote:

> On Tue, Dec 06, 2005 at 08:19:34PM +0200, Peter wrote:
>> You could introduce some metrics using f.ex. (power * known_life)/price,
>> otherwise it's comparing apples with prunes. A brand name PSU would have
>> to run 4 times longer than a no name cheap one on account of the price
>> difference alone. That is equivalent to roughly being twice as large or
>> running 20 degrees (Celsius) cooler (imagine the fans this requires,
>> given standard PSU size).
>
> You're assuming that both power supplies cause no damage when
> they fail.  If the no name supply destroys your motherboard
> or disk when it fails, then the brand name supply only has to
> last a small fraction of the time to be a bargain.  Even with
> a clean failure, if you put any value on having your computer
> available and on the cost of your time to get a replacement
> power supply and install it, the brand name supply is also
> still a cheaper deal if it lasts a little while longer.

Okay, my mistake, I omitted the important part: Real users who care 
replace the power supplies after a given number of hours estimated as 
above. Same for disks and same for the entire system (there are 
capacitors that wear out on the motherboard SMSPUs also - and I do not 
mean the fake ones that pop early). The trick is to know when problems 
are due and to have *sheduled* downtime as opposed to unsheduled. And 
the only thing that the PSU price changes is the time sheduled to the 
change. Of course a safe(r) way is to add a crowbar protector in the 
secondary (and not rely on the PSUs) - this is very basic reliability 
engineering - you do not rely on a failing part to do the safekeeping of 
other parts.

Every serious piece of equipment with life cycle management has such a 
policy in place. If you wait for it to fail you are inviting disaster. 
There never has been, and there never will be such a thing as a 
'perfectly safe' failure mode. Look at the big guys (NASA, ESA, you name 
it). When you rely on a safety mechanism to save something and you have 
the alternative of shutting it down beforehead you are, imho, playing 
with fire.

Hours meters are your best friends. smartctl -a etc will tell you a lot 
about a system (and its motherboard) assuming they were 'mated' at 
'birth'. By the time your no name PSU hits 2 years (~20k hours) start 
worrying. A good name PSU maximum 4 years. A disk 2 years (IDE). Any P3+ 
motherboard with smpsu on board will 'expire' in up to 3 years for the 
same reason. If the equipment runs hot (no a/c) the times are lower. 
These figures are based on limited experience, so use with some care.

Peter
--
The Toronto Linux Users Group.      Meetings: http://tlug.ss.org
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://tlug.ss.org/subscribe.shtml