(Sorry about the blank email... too much caffeene got me a twitchy
trigger finger).
Dimitri Maziuk wrote:
>
> In fact,
> there will be some point at which each individual PSU will run
> just as hot as if it handled all the load on its own (you can be sure
> your box will draw exactly that much load, thank you Mr Murphy).
>
Hmmm, I'm doubtful that that's the case if our load is around 25% to 50%
or something like that. That's based on almost a pure guess, but I
recall UPS efficiency numbers of something like high 90% numbers for
almost full load versus 85% or so at 50% load. Obviously, a very lightly
loaded power supply will put out about as much heat regardless of
whether it's running at 1% or 2% capacity, but I'm thinking that will
diverge fairly quickly.
Also, in most PS and case designs, the PS should have a negligable
effect on the heat transfer of the box, given that its fan exhausts
directly to the exterior of the case. If the case is ventilated through
the power supply (as common in desktop PCs and low end servers), there
will be a small change, depending on the temperature within the power
supply case (been about 4-years since I decided 'Thermo and an ME degree
weren't for me, so the ideas are fuzzy for me too).
> I'm not arguing that this is the case, I'm saying that this kind of
> argument can be twisted and turned any which way you like.
>
Very true, which is why redundancy will be the main factor, not heat
production.
> I had mid-80s AT PSUs. They'd still be working if I didn't have to
> move house and throw all that junk out. If a PSU lasts for 15 years,
> will 2 load-sharing PSUs last 30 years? Do I care? (Will I last 30
> more years?) I know that CPU will maybe last 1/5th of that, disks
> maybe 1/3rd. So what is it I'm going to achieve by setting up
> load-balancing PSUs?
>
Well, those mid-80s AT PSUs (in general, I mean) seem to have either
been A) oversized for the AT PCs or B) just better quality than normal
desktop-grade power supplies of today. (BTW: I recall hearing that Intel
speced Pentium CPUs for a lifetime of 10 or 20 years in normal usage...
used as a rationale for overclocking and the reduced lifespan it
causes). No, I don't anticipate a linear relationship between load and
lifespan, nor would I anticipate a linear relationship between load and
heat disappated, heat dissapated and lifespan, etc... I would however,
anticipate that keeping a power supply running somewhere under its full
rated capacity will increase its lifespan to some extent. Also, in a
load-balancing configuration, you eliminate the
> ... Also consider what
> > happens if the load was near the capacity of a single supply, and spiked
> > over the capacity. If we were using the second supply as a "backup" to
> > only be switched in if the primary failed, how would that be handled?
>
> Well if you mean some piece of hardware suddently decides to draw
> $BIGNUM times its normal current, the PSU will die. Depending on the
> design, there's a circuit somewhere (eg. on the backplane) that does
> the appropriate magic and switches the second PSU on. Of course it'll
> die very soon too, unless the FPOH in question magically fixed itself
> in the meantime.
>
I was thinking more like the combined load in the box was something like
95% of the rated capacity of the power supply, then spiked to 110% (like
having a bunch of SCSI drives spin up). A decent power supply probably
won't let the smoke out, but it probably won't give the best power
either. A redundant load-sharing arrangement would have both supplies
running at something like 42.5%, then spike to something like 55%.
Granted, this is a bit of a stretch, but I've seen too many cases
recently of servers in a simple consumer PC box that gradually got
stuffed full of SCSI drives until a PS failed.
> Sometimes the magic fails -- I remember the look on my boss's face when
> he pulled a hot-swappable PSU out of a live swerver, and the box went
> down. Oops. (Only happened once; we later tried to reproduce the problem,
> quite unsuccessfully: PSUs switched over like a charm, every bloody time.
> Surprise, surprise.)
>
> Dima
Like you said, Mr. Murphy pays a visit every now and then :-)
(whew... ok, I'm done.. this topic has wandered far enough!)
--Rich
--
_________________________________________________________
Rich Puhek
ETN Systems Inc.
_________________________________________________________
